refactor(ai): unify prompt cache configuration

Co-authored-by: Aiden Cline <rekram1-node@users.noreply.github.com>

bun.lock

@@ -67,6 +67,7 @@
         "@opencode-ai/sdk": "file:vendor/opencode-ai-sdk-1.18.8-dev.tgz",
         "@opencode-ai/session-ui": "workspace:*",
         "@opencode-ai/ui": "workspace:*",
+        "@opencode-ai/util": "workspace:*",
         "@pierre/trees": "1.0.0-beta.4",
         "@sentry/solid": "catalog:",
         "@shikijs/transformers": "3.9.2",
@@ -520,6 +521,7 @@
         "@opencode-ai/core": "workspace:*",
         "@opencode-ai/session-ui": "workspace:*",
         "@opencode-ai/ui": "workspace:*",
+        "@opencode-ai/util": "workspace:*",
         "@pierre/diffs": "catalog:",
         "@solidjs/meta": "catalog:",
         "@solidjs/router": "catalog:",

packages/ai/AGENTS.md

@@ -12,6 +12,8 @@
 
 Per-type constructors live on the type, not as top-level re-exports. Use `Message.system(...)`, `Message.user(...)`, `Message.assistant(...)`, `Message.tool(...)`, `Model.make(...)`, `ToolDefinition.make(...)`, `ToolCallPart.make(...)`, `ToolResultPart.make(...)`, `ToolChoice.make(...)`, `ToolChoice.named(...)`, `SystemPart.make(...)`, and `GenerationOptions.make(...)` directly. The top-level `LLM` namespace is reserved for request-shaped call APIs: `LLM.request`, `LLM.generate`, `LLM.stream`, `LLM.updateRequest`, and `LLM.generateObject`. Two ways to construct the same thing is one too many.
 
+- Keep provider-defined string enums forward-compatible. Expose known values for autocomplete while accepting future values with `Known | (string & {})`; use `Schema.String` at runtime unless rejecting unknown values is required for correctness.
+
 ## Tests
 
 - Use `testEffect(...)` from `test/lib/effect.ts` for tests requiring Effect layers.

packages/ai/example/tutorial.ts

@@ -33,9 +33,10 @@ const model = OpenAI.configure({
 //
 //   - `generation`: common controls such as max tokens, temperature, topP/topK,
 //     penalties, seed, and stop sequences.
+//   - `cache`: provider-neutral prompt caching behavior and cache affinity.
 //   - `providerOptions`: namespaced provider-native behavior. For example,
-//     OpenAI cache keys and store behavior, Anthropic thinking, Gemini thinking
-//     config, or OpenRouter routing/reasoning.
+//     OpenAI store behavior, Anthropic thinking, Gemini thinking config, or
+//     OpenRouter routing/reasoning.
 //   - `http`: last-resort serializable overlays for final request body, headers,
 //     and query params. Prefer typed `providerOptions` when a field is stable.
 //
@@ -45,9 +46,7 @@ const request = LLM.request({
   system: "You are concise and practical.",
   prompt: "Tell me a joke",
   generation: { maxTokens: 80, temperature: 0.7 },
-  providerOptions: {
-    openai: { promptCacheKey: "tutorial-joke" },
-  },
+  cache: { mode: "auto", key: "tutorial-joke" },
 })
 
 // 3. `generate` sends the request and collects the event stream into one

packages/ai/src/cache-policy.ts

@@ -10,16 +10,16 @@
 //
 // Manual `cache: CacheHint` placements on individual parts are preserved and
 // count against the four-breakpoint budget; auto only fills remaining slots.
-import { CacheHint, type CachePolicy, type CachePolicyObject } from "./schema/options"
+import { CacheHint, type CacheExplicit, type CachePolicy } from "./schema/options"
 import { LLMRequest, Message, ToolDefinition, type ContentPart } from "./schema/messages"
 
-const AUTO: CachePolicyObject = {
+const AUTO: Omit<CacheExplicit, "mode" | "key"> = {
   tools: true,
   system: true,
   messages: { tail: 1 },
 }
 
-const NONE: CachePolicyObject = {}
+const NONE: Omit<CacheExplicit, "mode" | "key"> = {}
 const BREAKPOINT_CAP = 4
 
 // Resolution rules:
@@ -29,8 +29,8 @@ const BREAKPOINT_CAP = 4
 //   - "auto"      → tools + first/last system + final message boundary.
 //   - "none"      → no auto placement; manual `CacheHint`s still flow.
 //   - object form → exactly what the caller asked for.
-const resolve = (policy: CachePolicy | undefined): CachePolicyObject => {
-  if (policy === undefined || policy === "auto") return AUTO
+const resolve = (policy: CachePolicy | undefined): Omit<CacheExplicit, "mode" | "key"> => {
+  if (policy === undefined || (policy !== "none" && policy.mode === "auto")) return AUTO
   if (policy === "none") return NONE
   return policy
 }
@@ -38,7 +38,7 @@ const resolve = (policy: CachePolicy | undefined): CachePolicyObject => {
 // Protocols whose wire format ignores inline cache markers (OpenAI's implicit
 // prefix caching, Gemini's implicit + out-of-band CachedContent). Skip the
 // whole policy pass for these — emitting hints would be harmless but pointless.
-const RESPECTS_INLINE_HINTS = new Set(["anthropic-messages", "bedrock-converse"])
+const RESPECTS_INLINE_HINTS = new Set(["anthropic-messages", "bedrock-converse", "openrouter"])
 
 const makeHint = (ttlSeconds: number | undefined): CacheHint =>
   ttlSeconds !== undefined ? new CacheHint({ type: "ephemeral", ttlSeconds }) : new CacheHint({ type: "ephemeral" })
@@ -103,7 +103,7 @@ const markMessageAt = (
 
 const markMessages = (
   messages: ReadonlyArray<Message>,
-  strategy: NonNullable<CachePolicyObject["messages"]>,
+  strategy: NonNullable<CacheExplicit["messages"]>,
   hint: CacheHint,
   budget: Budget,
 ): ReadonlyArray<Message> => {
@@ -133,6 +133,11 @@ const countHints = (request: LLMRequest) =>
 
 export const applyCachePolicy = (request: LLMRequest): LLMRequest => {
   if (!RESPECTS_INLINE_HINTS.has(request.model.route.id)) return request
+  if (
+    request.model.route.id === "openrouter" &&
+    (request.cache === undefined || (request.cache !== "none" && request.cache.mode === "auto"))
+  )
+    return request
   const policy = resolve(request.cache)
   if (!policy.tools && !policy.system && !policy.messages) return request
 

packages/ai/src/protocols/gemini.ts

@@ -29,9 +29,30 @@ export const DEFAULT_BASE_URL = "https://generativelanguage.googleapis.com/v1bet
 
 export interface OptionsInput {
   readonly [key: string]: unknown
+  readonly cachedContent?: string
+  readonly safetySettings?: ReadonlyArray<{
+    readonly category:
+      | "HARM_CATEGORY_UNSPECIFIED"
+      | "HARM_CATEGORY_HATE_SPEECH"
+      | "HARM_CATEGORY_DANGEROUS_CONTENT"
+      | "HARM_CATEGORY_HARASSMENT"
+      | "HARM_CATEGORY_SEXUALLY_EXPLICIT"
+      | "HARM_CATEGORY_CIVIC_INTEGRITY"
+      | (string & {})
+    readonly threshold:
+      | "HARM_BLOCK_THRESHOLD_UNSPECIFIED"
+      | "BLOCK_LOW_AND_ABOVE"
+      | "BLOCK_MEDIUM_AND_ABOVE"
+      | "BLOCK_ONLY_HIGH"
+      | "BLOCK_NONE"
+      | "OFF"
+      | (string & {})
+  }>
+  readonly serviceTier?: "standard" | "flex" | "priority" | (string & {})
   readonly thinkingConfig?: {
     readonly thinkingBudget?: number
     readonly includeThoughts?: boolean
+    readonly thinkingLevel?: "minimal" | "low" | "medium" | "high" | (string & {})
   }
 }
 
@@ -111,6 +132,12 @@ const GeminiToolConfig = Schema.Struct({
 const GeminiThinkingConfig = Schema.Struct({
   thinkingBudget: Schema.optional(Schema.Number),
   includeThoughts: Schema.optional(Schema.Boolean),
+  thinkingLevel: Schema.optional(Schema.String),
+})
+
+const GeminiSafetySetting = Schema.Struct({
+  category: Schema.String,
+  threshold: Schema.String,
 })
 
 const GeminiGenerationConfig = Schema.Struct({
@@ -123,7 +150,10 @@ const GeminiGenerationConfig = Schema.Struct({
 })
 
 const GeminiBodyFields = {
+  cachedContent: Schema.optional(Schema.String),
   contents: Schema.Array(GeminiContent),
+  safetySettings: optionalArray(GeminiSafetySetting),
+  serviceTier: Schema.optional(Schema.String),
   systemInstruction: Schema.optional(GeminiSystemInstruction),
   tools: optionalArray(GeminiTool),
   toolConfig: Schema.optional(GeminiToolConfig),
@@ -316,17 +346,38 @@ const lowerMessages = Effect.fn("Gemini.lowerMessages")(function* (request: LLMR
 })
 
 const resolveOptions = (request: LLMRequest) => {
-  const value = request.providerOptions?.gemini?.thinkingConfig
-  if (!ProviderShared.isRecord(value)) return {}
+  const input = request.providerOptions?.gemini
+  const value = input?.thinkingConfig
   const thinkingConfig = {
-    thinkingBudget: typeof value.thinkingBudget === "number" ? value.thinkingBudget : undefined,
-    includeThoughts: typeof value.includeThoughts === "boolean" ? value.includeThoughts : undefined,
+    thinkingBudget:
+      ProviderShared.isRecord(value) && typeof value.thinkingBudget === "number" ? value.thinkingBudget : undefined,
+    includeThoughts:
+      ProviderShared.isRecord(value) && typeof value.includeThoughts === "boolean"
+        ? value.includeThoughts
+        : ProviderShared.isRecord(value)
+          ? true
+          : undefined,
+    thinkingLevel:
+      ProviderShared.isRecord(value) && typeof value.thinkingLevel === "string" ? value.thinkingLevel : undefined,
   }
   return {
+    cachedContent: typeof input?.cachedContent === "string" ? input.cachedContent : undefined,
+    safetySettings: mapSafetySettings(input?.safetySettings),
+    serviceTier: typeof input?.serviceTier === "string" ? input.serviceTier : undefined,
     thinkingConfig: Object.values(thinkingConfig).some((item) => item !== undefined) ? thinkingConfig : undefined,
   }
 }
 
+function mapSafetySettings(value: unknown) {
+  if (!Array.isArray(value)) return undefined
+  const settings = value.flatMap((item) =>
+    ProviderShared.isRecord(item) && typeof item.category === "string" && typeof item.threshold === "string"
+      ? [{ category: item.category, threshold: item.threshold }]
+      : [],
+  )
+  return settings
+}
+
 const fromRequest = Effect.fn("Gemini.fromRequest")(function* (request: LLMRequest) {
   const hasTools = request.tools.length > 0
   const generation = request.generation
@@ -342,7 +393,10 @@ const fromRequest = Effect.fn("Gemini.fromRequest")(function* (request: LLMReque
   }
 
   return {
+    cachedContent: options.cachedContent,
     contents: yield* lowerMessages(request),
+    safetySettings: options.safetySettings,
+    serviceTier: options.serviceTier,
     systemInstruction:
       request.system.length === 0 ? undefined : { parts: [{ text: ProviderShared.joinText(request.system) }] },
     tools: hasTools

packages/ai/src/protocols/open-responses.ts

@@ -3,6 +3,7 @@ import type { Content } from "@opencode-ai/schema/tool"
 import { HttpTransport } from "../route/transport"
 import { Protocol } from "../route/protocol"
 import {
+  cacheKey,
   LLMError,
   LLMEvent,
   Usage,
@@ -542,7 +543,7 @@ const lowerOptions = (request: LLMRequest) => {
   return {
     ...(options.instructions ? { instructions: options.instructions } : {}),
     ...(options.store !== undefined ? { store: options.store } : {}),
-    ...(options.promptCacheKey ? { prompt_cache_key: options.promptCacheKey } : {}),
+    ...(cacheKey(request.cache) ? { prompt_cache_key: cacheKey(request.cache) } : {}),
     ...(options.include ? { include: options.include } : {}),
     ...(options.reasoningEffort || options.reasoningSummary
       ? { reasoning: { effort: options.reasoningEffort, summary: options.reasoningSummary } }

packages/ai/src/protocols/openai-chat.ts

@@ -11,6 +11,7 @@ import {
   Usage,
   type FinishReason,
   type FinishReasonDetails,
+  type CacheHint,
   type JsonSchema,
   type LLMRequest,
   type MediaPart,
@@ -38,6 +39,11 @@ export const PATH = "/chat/completions"
 // The body schema is the provider-native JSON body. `fromRequest` below builds
 // this shape from the common `LLMRequest`, then `Route.make` validates and
 // JSON-encodes it before transport.
+const OpenAIChatCacheControl = Schema.Struct({
+  type: Schema.Literal("ephemeral"),
+  ttl: Schema.optional(Schema.String),
+})
+
 const OpenAIChatFunction = Schema.Struct({
   name: Schema.String,
   description: Schema.String,
@@ -47,6 +53,7 @@ const OpenAIChatFunction = Schema.Struct({
 const OpenAIChatTool = Schema.Struct({
   type: Schema.tag("function"),
   function: OpenAIChatFunction,
+  cache_control: Schema.optional(OpenAIChatCacheControl),
 })
 type OpenAIChatTool = Schema.Schema.Type<typeof OpenAIChatTool>
 
@@ -61,7 +68,11 @@ const OpenAIChatAssistantToolCall = Schema.Struct({
 type OpenAIChatAssistantToolCall = Schema.Schema.Type<typeof OpenAIChatAssistantToolCall>
 
 const OpenAIChatUserContent = Schema.Union([
-  Schema.Struct({ type: Schema.Literal("text"), text: Schema.String }),
+  Schema.Struct({
+    type: Schema.Literal("text"),
+    text: Schema.String,
+    cache_control: Schema.optional(OpenAIChatCacheControl),
+  }),
   Schema.Struct({
     type: Schema.Literal("image_url"),
     image_url: Schema.Struct({ url: Schema.String }),
@@ -69,7 +80,10 @@ const OpenAIChatUserContent = Schema.Union([
 ])
 
 const OpenAIChatMessage = Schema.Union([
-  Schema.Struct({ role: Schema.Literal("system"), content: Schema.String }),
+  Schema.Struct({
+    role: Schema.Literal("system"),
+    content: Schema.Union([Schema.String, Schema.Array(OpenAIChatUserContent)]),
+  }),
   Schema.Struct({
     role: Schema.Literal("user"),
     content: Schema.Union([Schema.String, Schema.Array(OpenAIChatUserContent)]),
@@ -83,10 +97,16 @@ const OpenAIChatMessage = Schema.Union([
       reasoning: Schema.optional(Schema.String),
       reasoning_text: Schema.optional(Schema.String),
       reasoning_details: Schema.optional(Schema.Unknown),
+      cache_control: Schema.optional(OpenAIChatCacheControl),
     }),
     [Schema.Record(Schema.String, Schema.Unknown)],
   ),
-  Schema.Struct({ role: Schema.Literal("tool"), tool_call_id: Schema.String, content: Schema.String }),
+  Schema.Struct({
+    role: Schema.Literal("tool"),
+    tool_call_id: Schema.String,
+    content: Schema.String,
+    cache_control: Schema.optional(OpenAIChatCacheControl),
+  }),
 ]).pipe(Schema.toTaggedUnion("role"))
 type OpenAIChatMessage = Schema.Schema.Type<typeof OpenAIChatMessage>
 
@@ -107,6 +127,7 @@ export const bodyFields = {
   stream_options: Schema.optional(Schema.Struct({ include_usage: Schema.Boolean })),
   store: Schema.optional(Schema.Boolean),
   reasoning_effort: Schema.optional(OpenAIOptions.OpenAIReasoningEffort),
+  max_completion_tokens: Schema.optional(Schema.Number),
   max_tokens: Schema.optional(Schema.Number),
   temperature: Schema.optional(Schema.Number),
   top_p: Schema.optional(Schema.Number),
@@ -209,13 +230,20 @@ export interface ParserState {
 // Lowering is the only place that knows how common LLM messages map onto the
 // OpenAI Chat wire format. Keep provider quirks here instead of leaking native
 // fields into `LLMRequest`.
-const lowerTool = (tool: ToolDefinition, inputSchema: JsonSchema): OpenAIChatTool => ({
+interface LoweringOptions {
+  readonly cacheControl?: (
+    cache: CacheHint | undefined,
+  ) => Schema.Schema.Type<typeof OpenAIChatCacheControl> | undefined
+}
+
+const lowerTool = (tool: ToolDefinition, inputSchema: JsonSchema, options: LoweringOptions): OpenAIChatTool => ({
   type: "function",
   function: {
     name: tool.name,
     description: tool.description,
     parameters: ToolSchemaProjection.openAI(inputSchema),
   },
+  cache_control: options.cacheControl?.(tool.cache),
 })
 
 const lowerToolChoice = (toolChoice: NonNullable<LLMRequest["toolChoice"]>) =>
@@ -257,11 +285,14 @@ const reasoningDetails = (parts: ReadonlyArray<ReasoningPart>, native: unknown)
   if (isRecord(native) && Array.isArray(native.reasoning_details)) return native.reasoning_details
 }
 
-const lowerUserMessage = Effect.fn("OpenAIChat.lowerUserMessage")(function* (message: OpenAIChatRequestMessage) {
+const lowerUserMessage = Effect.fn("OpenAIChat.lowerUserMessage")(function* (
+  message: OpenAIChatRequestMessage,
+  options: LoweringOptions,
+) {
   const content: Array<Schema.Schema.Type<typeof OpenAIChatUserContent>> = []
   for (const part of message.content) {
     if (part.type === "text") {
-      content.push({ type: "text", text: part.text })
+      content.push({ type: "text", text: part.text, cache_control: options.cacheControl?.(part.cache) })
       continue
     }
     if (part.type === "media") {
@@ -270,14 +301,18 @@ const lowerUserMessage = Effect.fn("OpenAIChat.lowerUserMessage")(function* (mes
     }
     return yield* ProviderShared.unsupportedContent("OpenAI Chat", "user", ["text", "media"])
   }
-  if (content.every((part) => part.type === "text"))
-    return { role: "user" as const, content: content.map((part) => part.text).join("") }
+  if (content.every((part) => part.type === "text" && part.cache_control === undefined))
+    return {
+      role: "user" as const,
+      content: content.map((part) => (part.type === "text" ? part.text : "")).join(""),
+    }
   return { role: "user" as const, content }
 })
 
 const lowerAssistantMessage = Effect.fn("OpenAIChat.lowerAssistantMessage")(function* (
   message: OpenAIChatRequestMessage,
   configuredField?: string,
+  options: LoweringOptions = {},
 ) {
   const content: TextPart[] = []
   const reasoning: ReasoningPart[] = []
@@ -315,29 +350,44 @@ const lowerAssistantMessage = Effect.fn("OpenAIChat.lowerAssistantMessage")(func
     if (reasoning.length === 0) return nativeReasoning
     return text
   })()
+  const cached = message.content.findLast((part) => "cache" in part && part.cache !== undefined)
   const result = {
     role: "assistant" as const,
     content: content.length === 0 ? null : ProviderShared.joinText(content),
     tool_calls: toolCalls.length === 0 ? undefined : toolCalls,
     reasoning_details: details,
+    cache_control: options.cacheControl?.(cached && "cache" in cached ? cached.cache : undefined),
   }
   if (field === undefined || reasoningText === undefined) return result
   return { ...result, [field]: reasoningText }
 })
 
-const lowerToolMessages = Effect.fn("OpenAIChat.lowerToolMessages")(function* (message: OpenAIChatRequestMessage) {
+const lowerToolMessages = Effect.fn("OpenAIChat.lowerToolMessages")(function* (
+  message: OpenAIChatRequestMessage,
+  options: LoweringOptions,
+) {
   const messages: OpenAIChatMessage[] = []
   const images: Array<Schema.Schema.Type<typeof OpenAIChatUserContent>> = []
   for (const part of message.content) {
     if (!ProviderShared.supportsContent(part, ["tool-result"]))
       return yield* ProviderShared.unsupportedContent("OpenAI Chat", "tool", ["tool-result"])
     if (part.result.type !== "content") {
-      messages.push({ role: "tool", tool_call_id: part.id, content: ProviderShared.toolResultText(part) })
+      messages.push({
+        role: "tool",
+        tool_call_id: part.id,
+        content: ProviderShared.toolResultText(part),
+        cache_control: options.cacheControl?.(part.cache),
+      })
       continue
     }
     const content: ReadonlyArray<Tool.Content> = part.result.value
     const text = content.filter((item) => item.type === "text").map((item) => item.text)
-    messages.push({ role: "tool", tool_call_id: part.id, content: text.join("\n") })
+    messages.push({
+      role: "tool",
+      tool_call_id: part.id,
+      content: text.join("\n"),
+      cache_control: options.cacheControl?.(part.cache),
+    })
     const files = content.filter((item) => item.type === "file")
     images.push(
       ...(yield* Effect.forEach(files, (item) =>
@@ -351,15 +401,29 @@ const lowerToolMessages = Effect.fn("OpenAIChat.lowerToolMessages")(function* (m
 const lowerMessage = Effect.fn("OpenAIChat.lowerMessage")(function* (
   message: OpenAIChatRequestMessage,
   reasoningField?: string,
+  options: LoweringOptions = {},
 ) {
-  if (message.role === "user") return [yield* lowerUserMessage(message)]
-  if (message.role === "assistant") return [yield* lowerAssistantMessage(message, reasoningField)]
-  return (yield* lowerToolMessages(message)).messages
+  if (message.role === "user") return [yield* lowerUserMessage(message, options)]
+  if (message.role === "assistant") return [yield* lowerAssistantMessage(message, reasoningField, options)]
+  return (yield* lowerToolMessages(message, options)).messages
 })
 
-const lowerMessages = Effect.fn("OpenAIChat.lowerMessages")(function* (request: LLMRequest) {
+const lowerMessages = Effect.fn("OpenAIChat.lowerMessages")(function* (request: LLMRequest, options: LoweringOptions) {
   const system: OpenAIChatMessage[] =
-    request.system.length === 0 ? [] : [{ role: "system", content: ProviderShared.joinText(request.system) }]
+    request.system.length === 0
+      ? []
+      : request.system.some((part) => part.cache !== undefined) && options.cacheControl !== undefined
+        ? [
+            {
+              role: "system",
+              content: request.system.map((part) => ({
+                type: "text",
+                text: part.text,
+                cache_control: options.cacheControl?.(part.cache),
+              })),
+            },
+          ]
+        : [{ role: "system", content: ProviderShared.joinText(request.system) }]
   const messages = [...system]
   const pendingImages: Array<Schema.Schema.Type<typeof OpenAIChatUserContent>> = []
   const flushImages = () => {
@@ -370,28 +434,53 @@ const lowerMessages = Effect.fn("OpenAIChat.lowerMessages")(function* (request:
     if (message.role === "system") {
       const part = yield* ProviderShared.wrappedSystemUpdate("OpenAI Chat", message)
       if (pendingImages.length > 0) {
-        messages.push({ role: "user", content: [...pendingImages.splice(0), { type: "text", text: part.text }] })
+        messages.push({
+          role: "user",
+          content: [
+            ...pendingImages.splice(0),
+            { type: "text", text: part.text, cache_control: options.cacheControl?.(part.cache) },
+          ],
+        })
         continue
       }
       const previous = messages.at(-1)
       if (previous?.role === "user" && typeof previous.content === "string")
-        messages[messages.length - 1] = { role: "user", content: `${previous.content}\n${part.text}` }
+        messages[messages.length - 1] = options.cacheControl?.(part.cache)
+          ? {
+              role: "user",
+              content: [
+                { type: "text", text: previous.content },
+                { type: "text", text: part.text, cache_control: options.cacheControl(part.cache) },
+              ],
+            }
+          : { role: "user", content: `${previous.content}\n${part.text}` }
       else if (previous?.role === "user" && Array.isArray(previous.content))
         messages[messages.length - 1] = {
           role: "user",
-          content: [...previous.content, { type: "text", text: part.text }],
+          content: [
+            ...previous.content,
+            { type: "text", text: part.text, cache_control: options.cacheControl?.(part.cache) },
+          ],
         }
-      else messages.push({ role: "user", content: part.text })
+      else
+        messages.push(
+          options.cacheControl?.(part.cache)
+            ? {
+                role: "user",
+                content: [{ type: "text", text: part.text, cache_control: options.cacheControl(part.cache) }],
+              }
+            : { role: "user", content: part.text },
+        )
       continue
     }
     if (message.role === "tool") {
-      const lowered = yield* lowerToolMessages(message)
+      const lowered = yield* lowerToolMessages(message, options)
       messages.push(...lowered.messages)
       pendingImages.push(...lowered.images)
       continue
     }
     flushImages()
-    messages.push(...(yield* lowerMessage(message, request.model.compatibility?.reasoningField)))
+    messages.push(...(yield* lowerMessage(message, request.model.compatibility?.reasoningField, options)))
   }
   flushImages()
   return messages
@@ -405,7 +494,10 @@ const lowerOptions = (request: LLMRequest) => {
   }
 }
 
-const fromRequest = Effect.fn("OpenAIChat.fromRequest")(function* (request: LLMRequest) {
+export const fromRequest = Effect.fn("OpenAIChat.fromRequest")(function* (
+  request: LLMRequest,
+  options: LoweringOptions = {},
+) {
   // `fromRequest` returns the provider body only. Endpoint, auth, framing,
   // validation, and HTTP execution are composed by `Route.make`.
   const reasoningField = request.model.compatibility?.reasoningField
@@ -415,19 +507,26 @@ const fromRequest = Effect.fn("OpenAIChat.fromRequest")(function* (request: LLMR
     )
   const generation = request.generation
   const toolSchemaCompatibility = request.model.compatibility?.toolSchema
+  const maxTokensField = request.model.compatibility?.maxTokensField ?? "max_tokens"
   return {
     model: request.model.id,
-    messages: yield* lowerMessages(request),
+    messages: yield* lowerMessages(request, options),
     tools:
       request.tools.length === 0
         ? undefined
         : request.tools.map((tool) =>
-            lowerTool(tool, ToolSchemaProjection.modelCompatibility(tool.inputSchema, toolSchemaCompatibility)),
+            lowerTool(
+              tool,
+              ToolSchemaProjection.modelCompatibility(tool.inputSchema, toolSchemaCompatibility),
+              options,
+            ),
           ),
     tool_choice: request.toolChoice ? yield* lowerToolChoice(request.toolChoice) : undefined,
     stream: true as const,
     stream_options: { include_usage: true },
-    max_tokens: generation?.maxTokens,
+    ...(maxTokensField === "max_completion_tokens"
+      ? { max_completion_tokens: generation?.maxTokens }
+      : { max_tokens: generation?.maxTokens }),
     temperature: generation?.temperature,
     top_p: generation?.topP,
     frequency_penalty: generation?.frequencyPenalty,

packages/ai/src/protocols/utils/open-responses-options.ts

@@ -33,7 +33,6 @@ export const ServiceTierSchema = Schema.Literals(ServiceTiers)
 export interface Resolved {
   readonly instructions?: string
   readonly store?: boolean
-  readonly promptCacheKey?: string
   readonly reasoningEffort?: string
   readonly reasoningSummary?: "auto" | "concise" | "detailed"
   readonly include?: ReadonlyArray<ResponseIncludable>
@@ -50,7 +49,6 @@ export const resolve = (request: LLMRequest): Resolved => {
   return {
     instructions: typeof input?.instructions === "string" ? input.instructions : undefined,
     store: typeof input?.store === "boolean" ? input.store : undefined,
-    promptCacheKey: typeof input?.promptCacheKey === "string" ? input.promptCacheKey : undefined,
     reasoningEffort: typeof input?.reasoningEffort === "string" ? input.reasoningEffort : undefined,
     reasoningSummary:
       reasoningSummary === "auto" || reasoningSummary === "concise" || reasoningSummary === "detailed"

packages/ai/src/provider-package.ts

@@ -1,10 +1,12 @@
 import type { Model, ProviderOptions } from "./schema"
 
 export interface Settings extends Readonly<Record<string, unknown>> {
+  readonly baseURL?: string
   readonly headers?: Readonly<Record<string, string>>
   readonly body?: Readonly<Record<string, unknown>>
   readonly limits?: {
     readonly context: number
+    readonly input?: number
     readonly output: number
   }
 }

packages/ai/src/providers/open-responses-options.ts

@@ -5,7 +5,6 @@ export interface OpenResponsesOptionsInput {
   readonly [key: string]: unknown
   readonly instructions?: string
   readonly store?: boolean
-  readonly promptCacheKey?: string
   readonly reasoningEffort?: ReasoningEffort
   readonly reasoningSummary?: "auto" | "concise" | "detailed"
   readonly include?: ReadonlyArray<ResponseIncludable>

packages/ai/src/providers/openai-options.ts

@@ -17,7 +17,6 @@ const openAIProviderOptions = (options: OpenAIOptionsInput | undefined): Provide
   const openai = Object.fromEntries(
     definedEntries({
       store: options?.store,
-      promptCacheKey: options?.promptCacheKey,
       reasoningEffort: options?.reasoningEffort,
       reasoningSummary: options?.reasoningSummary,
       include: options?.include,

packages/ai/src/providers/openrouter.ts

@@ -4,21 +4,71 @@ import { Endpoint } from "../route/endpoint"
 import { Framing } from "../route/framing"
 import { Protocol } from "../route/protocol"
 import { AuthOptions, type ProviderAuthOption } from "../route/auth-options"
-import { ProviderID, type ModelID, type ProviderOptions } from "../schema"
+import { cacheKey, ProviderID, type CacheHint, type ModelID, type ProviderOptions } from "../schema"
 import type { ProviderPackage } from "../provider-package"
 import * as OpenAICompatibleProfiles from "./openai-compatible-profile"
 import * as OpenAIChat from "../protocols/openai-chat"
+import { newBreakpoints, ttlBucket } from "../protocols/utils/cache"
 import { isRecord } from "../protocols/shared"
 
 export const profile = OpenAICompatibleProfiles.profiles.openrouter
 export const id = ProviderID.make(profile.provider)
 const ADAPTER = "openrouter"
 
+type OpenRouterString<Known extends string> = Known | (string & {})
+
+export interface OpenRouterProviderRouting {
+  readonly [key: string]: unknown
+  readonly order?: ReadonlyArray<string>
+  readonly allow_fallbacks?: boolean
+  readonly require_parameters?: boolean
+  readonly data_collection?: OpenRouterString<"allow" | "deny">
+  readonly only?: ReadonlyArray<string>
+  readonly ignore?: ReadonlyArray<string>
+  readonly quantizations?: ReadonlyArray<string>
+  readonly sort?: OpenRouterString<"price" | "throughput" | "latency">
+  readonly max_price?: Readonly<{
+    prompt?: number | string
+    completion?: number | string
+    image?: number | string
+    audio?: number | string
+    request?: number | string
+  }>
+  readonly zdr?: boolean
+}
+
+export type OpenRouterPlugin =
+  | Readonly<{
+      id: "web"
+      max_results?: number
+      search_prompt?: string
+      engine?: OpenRouterString<"native" | "exa">
+    }>
+  | Readonly<{ id: "file-parser"; max_files?: number; pdf?: { engine?: string } }>
+  | Readonly<{ id: "moderation" }>
+  | Readonly<{ id: "response-healing" }>
+  | Readonly<{ id: "auto-router"; allowed_models?: ReadonlyArray<string> }>
+  | Readonly<{ id: string & {}; [key: string]: unknown }>
+
 export interface OpenRouterOptions {
   readonly [key: string]: unknown
-  readonly usage?: boolean | Record<string, unknown>
-  readonly reasoning?: Record<string, unknown>
-  readonly promptCacheKey?: string
+  readonly debug?: Readonly<{ echo_upstream_body?: boolean }>
+  readonly models?: ReadonlyArray<string>
+  readonly plugins?: ReadonlyArray<OpenRouterPlugin>
+  readonly provider?: OpenRouterProviderRouting
+  readonly reasoning?: Readonly<{
+    enabled?: boolean
+    exclude?: boolean
+    effort?: OpenRouterString<"none" | "minimal" | "low" | "medium" | "high" | "xhigh" | "max">
+    max_tokens?: number
+  }>
+  readonly usage?: boolean | Readonly<{ include: boolean }>
+  readonly user?: string
+  readonly web_search_options?: Readonly<{
+    max_results?: number
+    search_prompt?: string
+    engine?: OpenRouterString<"native" | "exa">
+  }>
 }
 
 export type OpenRouterProviderOptionsInput = ProviderOptions & {
@@ -47,7 +97,7 @@ export const protocol = Protocol.make({
   body: {
     schema: OpenRouterBody,
     from: (request) =>
-      OpenAIChat.protocol.body.from(request).pipe(
+      OpenAIChat.fromRequest(request, { cacheControl: cacheControl() }).pipe(
         Effect.map((body) => {
           const sourceAssistants = request.messages.filter((message) => message.role === "assistant")
           let assistantIndex = 0
@@ -71,6 +121,7 @@ export const protocol = Protocol.make({
             ...body,
             messages,
             ...bodyOptions(request.providerOptions?.openrouter),
+            ...(cacheKey(request.cache) ? { prompt_cache_key: cacheKey(request.cache) } : {}),
           } as OpenRouterBody
         }),
       ),
@@ -78,16 +129,38 @@ export const protocol = Protocol.make({
   stream: OpenAIChat.protocol.stream,
 })
 
+const cacheControl = () => {
+  const breakpoints = newBreakpoints(4)
+  return (cache: CacheHint | undefined) => {
+    if (cache === undefined || breakpoints.remaining === 0) return undefined
+    breakpoints.remaining -= 1
+    return {
+      type: "ephemeral" as const,
+      ...(ttlBucket(cache.ttlSeconds) === "1h" ? { ttl: "1h" } : {}),
+    }
+  }
+}
+
 const bodyOptions = (input: unknown) => {
   const openrouter = isRecord(input) ? input : {}
+  const { usage, models, provider, plugins, web_search_options, debug, user, reasoning, promptCacheKey, ...options } =
+    openrouter
   return {
-    ...(openrouter.usage === true
+    ...options,
+    ...(usage === undefined || usage === true
       ? { usage: { include: true } }
-      : isRecord(openrouter.usage)
-        ? { usage: openrouter.usage }
-        : {}),
-    ...(isRecord(openrouter.reasoning) ? { reasoning: openrouter.reasoning } : {}),
-    ...(typeof openrouter.promptCacheKey === "string" ? { prompt_cache_key: openrouter.promptCacheKey } : {}),
+      : usage === false
+        ? { usage: { include: false } }
+        : isRecord(usage)
+          ? { usage }
+          : {}),
+    ...(Array.isArray(models) ? { models } : {}),
+    ...(isRecord(provider) ? { provider } : {}),
+    ...(Array.isArray(plugins) ? { plugins } : {}),
+    ...(isRecord(web_search_options) ? { web_search_options } : {}),
+    ...(isRecord(debug) ? { debug } : {}),
+    ...(typeof user === "string" ? { user } : {}),
+    ...(isRecord(reasoning) ? { reasoning } : {}),
   }
 }
 

packages/ai/src/route/client.ts

@@ -5,7 +5,7 @@ import { Endpoint, type EndpointPatch } from "./endpoint"
 import { RequestExecutor } from "./executor"
 import { Framing } from "./framing"
 import { HttpTransport } from "./transport"
-import type { Transport, TransportRuntime } from "./transport"
+import type { HttpRequestTransform, Transport, TransportRuntime } from "./transport"
 import { WebSocketExecutor } from "./transport"
 import type { Protocol } from "./protocol"
 import { applyCachePolicy } from "../cache-policy"
@@ -46,7 +46,11 @@ export interface Route<Body, Prepared = unknown> {
   readonly body: RouteBody<Body>
   readonly with: (patch: RoutePatch<Body, Prepared>) => Route<Body, Prepared>
   readonly model: <Options extends ProviderOptions = ProviderOptions>(input: RouteMappedModelInput) => Model<Options>
-  readonly prepareTransport: (body: Body, request: LLMRequest) => Effect.Effect<Prepared, LLMError>
+  readonly prepareTransport: (
+    body: Body,
+    request: LLMRequest,
+    options?: StreamOptions,
+  ) => Effect.Effect<Prepared, LLMError>
   readonly streamPrepared: (
     prepared: Prepared,
     request: LLMRequest,
@@ -145,12 +149,16 @@ export interface Interface {
   readonly generate: GenerateMethod
 }
 
+export interface StreamOptions {
+  readonly transform?: HttpRequestTransform
+}
+
 export interface StreamMethod {
-  (request: LLMRequest): Stream.Stream<LLMEvent, LLMError>
+  (request: LLMRequest, options?: StreamOptions): Stream.Stream<LLMEvent, LLMError>
 }
 
 export interface GenerateMethod {
-  (request: LLMRequest): Effect.Effect<LLMResponse, LLMError>
+  (request: LLMRequest, options?: StreamOptions): Effect.Effect<LLMResponse, LLMError>
 }
 
 export class Service extends Context.Service<Service, Interface>()("@opencode/LLMClient") {}
@@ -286,7 +294,7 @@ function makeFromTransport<Body, Prepared, Frame, Event, State>(
       },
       model: <Options extends ProviderOptions = ProviderOptions>(input: RouteMappedModelInput) =>
         makeRouteModel<Options>(route, input),
-      prepareTransport: (body, request) =>
+      prepareTransport: (body, request, options) =>
         routeInput.transport.prepare({
           body,
           request,
@@ -294,6 +302,7 @@ function makeFromTransport<Body, Prepared, Frame, Event, State>(
           auth: routeInput.auth ?? Auth.none,
           encodeBody,
           headers: routeInput.headers,
+          transform: options?.transform,
         }),
       streamPrepared: (prepared: Prepared, request: LLMRequest, runtime: TransportRuntime) => {
         const route = `${request.model.provider}/${request.model.route.id}`
@@ -359,14 +368,14 @@ export function make<Body, Prepared, Frame, Event, State>(
   })
 }
 
-const compile = Effect.fn("LLM.compile")(function* (request: LLMRequest) {
+const compile = Effect.fn("LLM.compile")(function* (request: LLMRequest, options?: StreamOptions) {
   const resolved = applyCachePolicy(resolveRequestOptions(request))
   const route = resolved.model.route
 
   const body = yield* route.body
     .from(resolved)
     .pipe(Effect.flatMap(ProviderShared.validateWith(Schema.decodeUnknownEffect(route.body.schema))))
-  const prepared = yield* route.prepareTransport(body, resolved)
+  const prepared = yield* route.prepareTransport(body, resolved, options)
 
   return {
     request: resolved,
@@ -389,17 +398,17 @@ export const compileRequest = Effect.fn("LLM.compileRequest")(function* (request
   }
 })
 
-const streamRequestWith = (runtime: TransportRuntime) => (request: LLMRequest) =>
+const streamRequestWith = (runtime: TransportRuntime) => (request: LLMRequest, options?: StreamOptions) =>
   Stream.unwrap(
     Effect.gen(function* () {
-      const compiled = yield* compile(request)
+      const compiled = yield* compile(request, options)
       return compiled.route.streamPrepared(compiled.prepared, compiled.request, runtime)
     }),
   )
 
 const generateWith = (stream: Interface["stream"]) =>
-  Effect.fn("LLM.generate")(function* (request: LLMRequest) {
-    const state = yield* stream(request).pipe(Stream.runFold(LLMResponse.empty, LLMResponse.reduce))
+  Effect.fn("LLM.generate")(function* (request: LLMRequest, options?: StreamOptions) {
+    const state = yield* stream(request, options).pipe(Stream.runFold(LLMResponse.empty, LLMResponse.reduce))
     const response = LLMResponse.complete(state)
     if (response) return response
     return yield* ProviderShared.eventError(
@@ -408,24 +417,24 @@ const generateWith = (stream: Interface["stream"]) =>
     )
   })
 
-export function stream(request: LLMRequest): Stream.Stream<LLMEvent, LLMError> {
+export function stream(request: LLMRequest, options?: StreamOptions): Stream.Stream<LLMEvent, LLMError> {
   return Stream.unwrap(
     Effect.gen(function* () {
-      return (yield* Service).stream(request)
+      return (yield* Service).stream(request, options)
     }),
   ) as Stream.Stream<LLMEvent, LLMError>
 }
 
-export function generate(request: LLMRequest): Effect.Effect<LLMResponse, LLMError> {
+export function generate(request: LLMRequest, options?: StreamOptions): Effect.Effect<LLMResponse, LLMError> {
   return Effect.gen(function* () {
-    return yield* (yield* Service).generate(request)
+    return yield* (yield* Service).generate(request, options)
   }) as Effect.Effect<LLMResponse, LLMError>
 }
 
-export const streamRequest = (request: LLMRequest) =>
+export const streamRequest = (request: LLMRequest, options?: StreamOptions) =>
   Stream.unwrap(
     Effect.gen(function* () {
-      return (yield* Service).stream(request)
+      return (yield* Service).stream(request, options)
     }),
   )
 

packages/ai/src/route/index.ts

@@ -8,6 +8,7 @@ export type {
   AnyRoute,
   Interface as LLMClientShape,
   Service as LLMClientService,
+  StreamOptions,
 } from "./client"
 export * from "./executor"
 export { Auth } from "./auth"
@@ -22,4 +23,4 @@ export type { ApiKeyMode, AuthOverride, ProviderAuthOption } from "./auth-option
 export type { Definition as EndpointFn, EndpointInput } from "./endpoint"
 export type { Definition as FramingDef } from "./framing"
 export type { Protocol as ProtocolDef } from "./protocol"
-export type { Transport as TransportDef, TransportRuntime } from "./transport"
+export type { HttpRequest, HttpRequestTransform, Transport as TransportDef, TransportRuntime } from "./transport"

packages/ai/src/route/transport/http.ts

@@ -28,57 +28,9 @@ const applyQuery = (url: string, query: Record<string, string> | undefined) => {
   return next.toString()
 }
 
-const PROTOCOL_BODY_OVERLAY_DENYLIST = new Set([
-  "anthropic_version",
-  "content",
-  "contents",
-  "frequencyPenalty",
-  "frequency_penalty",
-  "generationConfig",
-  "inferenceConfig",
-  "input",
-  "maxTokens",
-  "max_tokens",
-  "messages",
-  "model",
-  "presencePenalty",
-  "presence_penalty",
-  "responseFormat",
-  "response_format",
-  "seed",
-  "stop",
-  "stopSequences",
-  "stop_sequences",
-  "stream",
-  "streamOptions",
-  "stream_options",
-  "system",
-  "systemInstruction",
-  "system_instruction",
-  "temperature",
-  "thinking",
-  "toolChoice",
-  "toolConfig",
-  "tool_choice",
-  "tool_config",
-  "tools",
-  "topK",
-  "topP",
-  "top_k",
-  "top_p",
-])
-
-const forbiddenBodyOverlayKeys = (body: Record<string, unknown>) =>
-  Object.keys(body).filter((key) => PROTOCOL_BODY_OVERLAY_DENYLIST.has(key))
-
 const bodyWithOverlay = <Body>(body: Body, request: LLMRequest, encodeBody: (body: Body) => string) =>
   Effect.gen(function* () {
     if (request.http?.body === undefined) return { jsonBody: body, bodyText: encodeBody(body) }
-    const forbiddenKeys = forbiddenBodyOverlayKeys(request.http.body)
-    if (forbiddenKeys.length > 0)
-      return yield* ProviderShared.invalidRequest(
-        `http.body cannot overlay protocol-owned field(s): ${forbiddenKeys.join(", ")}`,
-      )
     if (ProviderShared.isRecord(body)) {
       const overlaid = mergeJsonRecords(body, request.http.body) ?? {}
       return { jsonBody: overlaid, bodyText: ProviderShared.encodeJson(overlaid) }
@@ -120,14 +72,19 @@ export const httpJson = <Body, Frame>(input: HttpJsonInput<Body, Frame>): HttpJs
   id: "http-json",
   with: (patch) => httpJson({ ...input, ...patch }),
   prepare: (prepareInput) =>
-    jsonRequestParts({
-      ...prepareInput,
-    }).pipe(
-      Effect.map((parts) => ({
-        request: ProviderShared.jsonPost({ url: parts.url, body: parts.bodyText, headers: parts.headers }),
+    Effect.gen(function* () {
+      const parts = yield* jsonRequestParts({ ...prepareInput })
+      const request = { url: parts.url, method: "POST", headers: { ...parts.headers }, body: parts.bodyText }
+      yield* (prepareInput.transform?.(request) ?? Effect.void)
+      return {
+        request: ProviderShared.jsonPost({
+          url: request.url,
+          body: request.body ?? "",
+          headers: Headers.fromInput(request.headers),
+        }),
         framing: input.framing,
-      })),
-    ),
+      }
+    }),
   frames: (prepared, request, runtime) =>
     Stream.unwrap(
       runtime.http

packages/ai/src/route/transport/index.ts

@@ -10,6 +10,15 @@ export interface TransportRuntime {
   readonly webSocket?: WebSocketExecutorInterface
 }
 
+export interface HttpRequest {
+  url: string
+  readonly method: string
+  headers: Record<string, string>
+  body: string | undefined
+}
+
+export type HttpRequestTransform = (request: HttpRequest) => Effect.Effect<void>
+
 export interface Transport<Body, Prepared, Frame> {
   readonly id: string
   readonly prepare: (input: TransportPrepareInput<Body>) => Effect.Effect<Prepared, LLMError>
@@ -27,6 +36,7 @@ export interface TransportPrepareInput<Body> {
   readonly auth: Auth.Definition
   readonly encodeBody: (body: Body) => string
   readonly headers?: (input: { readonly request: LLMRequest }) => Record<string, string>
+  readonly transform?: HttpRequestTransform
 }
 
 export * as HttpTransport from "./http"

packages/ai/src/schema/messages.ts

@@ -124,6 +124,7 @@ export const ToolCallPart = Object.assign(
     name: Schema.String,
     input: Schema.Unknown,
     providerExecuted: Schema.optional(Schema.Boolean),
+    cache: Schema.optional(CacheHint),
     metadata: Schema.optional(Schema.Record(Schema.String, Schema.Unknown)),
     providerMetadata: Schema.optional(ProviderMetadata),
   }).annotate({ identifier: "LLM.Content.ToolCall" }),
@@ -168,6 +169,7 @@ export const ReasoningPart = Schema.Struct({
   type: Schema.Literal("reasoning"),
   text: Schema.String,
   encrypted: Schema.optional(Schema.String),
+  cache: Schema.optional(CacheHint),
   metadata: Schema.optional(Schema.Record(Schema.String, Schema.Unknown)),
   providerMetadata: Schema.optional(ProviderMetadata),
 }).annotate({ identifier: "LLM.Content.Reasoning" })

packages/ai/src/schema/options.ts

@@ -123,6 +123,7 @@ export const mergeGenerationOptions = (...items: ReadonlyArray<GenerationOptions
 
 export class ModelLimits extends Schema.Class<ModelLimits>("LLM.ModelLimits")({
   context: Schema.optional(Schema.Number),
+  input: Schema.optional(Schema.Number),
   output: Schema.optional(Schema.Number),
 }) {}
 
@@ -166,9 +167,13 @@ export namespace ModelDefaults {
 export const ModelToolSchemaCompatibility = Schema.Literals(["gemini", "moonshot"])
 export type ModelToolSchemaCompatibility = Schema.Schema.Type<typeof ModelToolSchemaCompatibility>
 
+export const ModelMaxTokensFieldCompatibility = Schema.Literals(["max_completion_tokens", "max_tokens"])
+export type ModelMaxTokensFieldCompatibility = Schema.Schema.Type<typeof ModelMaxTokensFieldCompatibility>
+
 export class ModelCompatibility extends Schema.Class<ModelCompatibility>("LLM.ModelCompatibility")({
   toolSchema: Schema.optional(ModelToolSchemaCompatibility),
   reasoningField: Schema.optional(Schema.String),
+  maxTokensField: Schema.optional(ModelMaxTokensFieldCompatibility),
 }) {}
 
 export namespace ModelCompatibility {
@@ -251,18 +256,17 @@ export class CacheHint extends Schema.Class<CacheHint>("LLM.CacheHint")({
   ttlSeconds: Schema.optional(Schema.Number),
 }) {}
 
-// Auto-placement policy for prompt caching. The protocol-neutral lowering step
-// reads this and injects `CacheHint`s at the configured boundaries; the
-// per-protocol body builders then translate those hints into wire markers as
-// usual. `"auto"` is the recommended default for agent loops — it places
-// breakpoints at the last tool definition, the first and last distinct system
-// parts, and the conversation tail. The rolling message breakpoint keeps a
-// prior cache entry within Anthropic/Bedrock's 20-block lookback during long
-// tool loops.
-//
-// Pass `"none"` to opt out entirely (the legacy behavior). Pass the granular
-// object form to override individual choices.
-export const CachePolicyObject = Schema.Struct({
+const CacheKey = { key: Schema.optional(Schema.String) }
+
+export const CacheAuto = Schema.Struct({
+  mode: Schema.Literal("auto"),
+  ...CacheKey,
+})
+export type CacheAuto = Schema.Schema.Type<typeof CacheAuto>
+
+export const CacheExplicit = Schema.Struct({
+  mode: Schema.Literal("explicit"),
+  ...CacheKey,
   tools: Schema.optional(Schema.Boolean),
   system: Schema.optional(Schema.Boolean),
   messages: Schema.optional(
@@ -274,7 +278,12 @@ export const CachePolicyObject = Schema.Struct({
   ),
   ttlSeconds: Schema.optional(Schema.Number),
 })
-export type CachePolicyObject = Schema.Schema.Type<typeof CachePolicyObject>
+export type CacheExplicit = Schema.Schema.Type<typeof CacheExplicit>
 
-export const CachePolicy = Schema.Union([Schema.Literal("auto"), Schema.Literal("none"), CachePolicyObject])
+// Omitted configuration uses automatic provider behavior and OpenCode's
+// automatic breakpoint placement where required. `"none"` sends no cache key
+// or explicit controls; providers may still cache implicitly.
+export const CachePolicy = Schema.Union([Schema.Literal("none"), CacheAuto, CacheExplicit])
 export type CachePolicy = Schema.Schema.Type<typeof CachePolicy>
+
+export const cacheKey = (cache: CachePolicy | undefined) => (cache && cache !== "none" ? cache.key : undefined)

packages/ai/test/cache-policy.test.ts

@@ -64,7 +64,7 @@ describe("applyCachePolicy", () => {
             Message.assistant("assistant reply"),
             Message.user("latest user message"),
           ],
-          cache: "auto",
+          cache: { mode: "auto" },
         }),
       )
 
@@ -93,7 +93,7 @@ describe("applyCachePolicy", () => {
           model: openaiModel,
           system: "Sys",
           prompt: "hi",
-          cache: "auto",
+          cache: { mode: "auto" },
         }),
       )
 
@@ -112,7 +112,7 @@ describe("applyCachePolicy", () => {
           model: geminiModel,
           system: "Sys",
           prompt: "hi",
-          cache: "auto",
+          cache: { mode: "auto" },
         }),
       )
 
@@ -133,7 +133,7 @@ describe("applyCachePolicy", () => {
           ],
           tools: [{ name: "t1", description: "t1", inputSchema: { type: "object", properties: {} } }],
           messages: [Message.user("first user"), Message.assistant("reply"), Message.user("latest user")],
-          cache: "auto",
+          cache: { mode: "auto" },
         }),
       )
 
@@ -183,7 +183,7 @@ describe("applyCachePolicy", () => {
           system: "Sys",
           tools: [{ name: "t1", description: "t1", inputSchema: { type: "object", properties: {} } }],
           prompt: "hi",
-          cache: { tools: true },
+          cache: { mode: "explicit", tools: true },
         }),
       )
 
@@ -204,7 +204,7 @@ describe("applyCachePolicy", () => {
             { type: "text", text: "last system" },
           ],
           prompt: "hi",
-          cache: "auto",
+          cache: { mode: "auto" },
         }),
       )
 
@@ -233,7 +233,7 @@ describe("applyCachePolicy", () => {
         ],
         tools: [{ name: "t1", description: "t1", inputSchema: { type: "object", properties: {} } }],
         prompt: "hi",
-        cache: "auto",
+          cache: { mode: "auto" },
       })
       const applied = applyCachePolicy(request)
       expect(applied.tools[0]?.cache).toBeDefined()
@@ -267,7 +267,7 @@ describe("applyCachePolicy", () => {
           model: anthropicModel,
           system: "Sys",
           prompt: "hi",
-          cache: { system: true, ttlSeconds: 3600 },
+          cache: { mode: "explicit", system: true, ttlSeconds: 3600 },
         }),
       )
 
@@ -283,7 +283,7 @@ describe("applyCachePolicy", () => {
         LLM.request({
           model: anthropicModel,
           messages: [Message.user("u1"), Message.assistant("a1"), Message.user("u2"), Message.assistant("a2")],
-          cache: { messages: { tail: 2 } },
+          cache: { mode: "explicit", messages: { tail: 2 } },
         }),
       )
 
@@ -301,7 +301,7 @@ describe("applyCachePolicy", () => {
         LLM.request({
           model: anthropicModel,
           messages: [Message.user("u1"), Message.assistant("a1"), Message.user("u2")],
-          cache: { messages: "latest-assistant" },
+          cache: { mode: "explicit", messages: "latest-assistant" },
         }),
       )
 

packages/ai/test/compile.test.ts

@@ -137,24 +137,61 @@ describe("request option precedence", () => {
     ),
   )
 
-  it.effect("rejects raw body overlays for protocol-owned roots", () =>
-    Effect.gen(function* () {
-      const model = OpenAIChat.route
-        .with({ endpoint: { baseURL: "https://api.openai.test/v1/" }, auth: Auth.bearer("test") })
-        .model({ id: "gpt-4o-mini" })
-      const error = yield* compileRequest(
-        LLM.request({
-          model,
-          prompt: "Say hello.",
-          http: { body: { model: "gpt-5", messages: [], tools: [] } },
-        }),
-      ).pipe(Effect.flip)
+  it.effect("transforms the final HTTP request after serialization and authentication", () =>
+    LLMClient.generate(
+      LLM.request({
+        model: OpenAIChat.route
+          .with({ endpoint: { baseURL: "https://api.openai.test/v1/" }, auth: Auth.bearer("fresh-key") })
+          .model({ id: "gpt-4o-mini" }),
+        prompt: "Say hello.",
+      }),
+      {
+        transform: (request) =>
+          Effect.sync(() => {
+            expect(request.headers.authorization).toBe("Bearer fresh-key")
+            request.url = "https://proxy.test/v1/chat/completions"
+            request.headers["x-plugin"] = "transformed"
+            request.body = JSON.stringify({ transformed: true })
+          }),
+      },
+    ).pipe(
+      Effect.provide(
+        dynamicResponse((input) =>
+          Effect.gen(function* () {
+            const web = yield* HttpClientRequest.toWeb(input.request).pipe(Effect.orDie)
+            expect(web.url).toBe("https://proxy.test/v1/chat/completions")
+            expect(web.headers.get("x-plugin")).toBe("transformed")
+            expect(decodeJson(input.text)).toEqual({ transformed: true })
+            return input.respond(sseEvents(deltaChunk({}, "stop")), {
+              headers: { "content-type": "text/event-stream" },
+            })
+          }),
+        ),
+      ),
+    ),
+  )
 
-      expect(error.reason).toMatchObject({
-        _tag: "InvalidRequest",
-        message: "http.body cannot overlay protocol-owned field(s): model, messages, tools",
-      })
-    }),
+  it.effect("applies raw body overlays after protocol lowering", () =>
+    LLMClient.generate(
+      LLM.request({
+        model: OpenAIChat.route
+          .with({ endpoint: { baseURL: "https://api.openai.test/v1/" }, auth: Auth.bearer("test") })
+          .model({ id: "gpt-4o-mini" }),
+        prompt: "Say hello.",
+        http: { body: { model: "gpt-5", messages: [], tools: [] } },
+      }),
+    ).pipe(
+      Effect.provide(
+        dynamicResponse((input) =>
+          Effect.gen(function* () {
+            expect(decodeJson(input.text)).toMatchObject({ model: "gpt-5", messages: [], tools: [] })
+            return input.respond(sseEvents(deltaChunk({}, "stop")), {
+              headers: { "content-type": "text/event-stream" },
+            })
+          }),
+        ),
+      ),
+    ),
   )
 
   it.effect("uses model output limits after route limits and before call maxTokens", () =>

packages/ai/test/provider-options/cloudflare.types.ts

@@ -3,11 +3,11 @@ import { CloudflareWorkersAI } from "../../src/providers"
 
 const model = CloudflareWorkersAI.configure({ accountId: "account", apiKey: "test" }).model("model")
 
-LLM.request({ model, prompt: "Hello", providerOptions: { openai: { promptCacheKey: "cache" } } })
+LLM.request({ model, prompt: "Hello", cache: { mode: "auto", key: "cache" } })
 
 LLM.request({
   model,
   prompt: "Hello",
-  // @ts-expect-error Cloudflare's OpenAI-compatible prompt cache key must be a string.
-  providerOptions: { openai: { promptCacheKey: 1 } },
+  // @ts-expect-error Prompt cache keys must be strings.
+  cache: { mode: "auto", key: 1 },
 })

packages/ai/test/provider-options/google.types.ts

@@ -9,9 +9,39 @@ LLM.request({
   providerOptions: { gemini: { thinkingConfig: { thinkingBudget: 1024 } } },
 })
 
+LLM.request({
+  model,
+  prompt: "Hello",
+  providerOptions: {
+    gemini: {
+      // @ts-expect-error Gemini safety settings require a threshold for every category.
+      safetySettings: [{ category: "HARM_CATEGORY_HATE_SPEECH" }],
+    },
+  },
+})
+
+LLM.request({
+  model,
+  prompt: "Hello",
+  providerOptions: {
+    gemini: {
+      cachedContent: "cachedContents/example",
+      safetySettings: [{ category: "HARM_CATEGORY_HATE_SPEECH", threshold: "BLOCK_ONLY_HIGH" }],
+      serviceTier: "future-tier",
+      thinkingConfig: { thinkingLevel: "high", includeThoughts: true },
+    },
+  },
+})
+
 LLM.request({
   model,
   prompt: "Hello",
   // @ts-expect-error Gemini thinking budgets must be numeric.
   providerOptions: { gemini: { thinkingConfig: { thinkingBudget: "large" } } },
 })
+
+LLM.request({
+  model,
+  prompt: "Hello",
+  providerOptions: { gemini: { thinkingConfig: { thinkingLevel: "maximum" } } },
+})

packages/ai/test/provider-options/openrouter.types.ts

@@ -5,6 +5,28 @@ const model = OpenRouter.provider.model("anthropic/claude-sonnet-4.5")
 
 LLM.request({ model, prompt: "Hello", providerOptions: { openrouter: { usage: true } } })
 
+LLM.request({
+  model,
+  prompt: "Hello",
+  providerOptions: {
+    openrouter: {
+      models: ["google/gemini-3.1-pro"],
+      provider: {
+        order: ["anthropic"],
+        require_parameters: true,
+        data_collection: "future-policy",
+        sort: "future-sort",
+        max_price: { prompt: "0.50" },
+      },
+      reasoning: { effort: "future-effort", exclude: false },
+      plugins: [{ id: "future-plugin", enabled: true }],
+      web_search_options: { engine: "future-engine" },
+      debug: { echo_upstream_body: true },
+      user: "user_123",
+    },
+  },
+})
+
 LLM.request({
   model,
   prompt: "Hello",

packages/ai/test/provider/gemini.test.ts

@@ -41,7 +41,14 @@ describe("Gemini route", () => {
     Effect.gen(function* () {
       const prepared = yield* compileRequest(
         LLMRequest.update(request, {
-          providerOptions: { gemini: { thinkingConfig: { thinkingBudget: 0, includeThoughts: false } } },
+          providerOptions: {
+            gemini: {
+              cachedContent: "cachedContents/example",
+              safetySettings: [{ category: "HARM_CATEGORY_HATE_SPEECH", threshold: "BLOCK_ONLY_HIGH" }],
+              serviceTier: "priority",
+              thinkingConfig: { thinkingBudget: 0, includeThoughts: false, thinkingLevel: "high" },
+            },
+          },
         }),
       )
       const filtered = yield* compileRequest(
@@ -49,12 +56,33 @@ describe("Gemini route", () => {
           providerOptions: { gemini: { thinkingConfig: { thinkingBudget: "invalid", includeThoughts: false } } },
         }),
       )
+      const defaulted = yield* compileRequest(
+        LLMRequest.update(request, {
+          providerOptions: { gemini: { thinkingConfig: { thinkingLevel: "high" } } },
+        }),
+      )
+      const emptySafetySettings = yield* compileRequest(
+        LLMRequest.update(request, {
+          providerOptions: { gemini: { safetySettings: [] } },
+        }),
+      )
 
       expect(prepared.body.generationConfig?.thinkingConfig).toEqual({
         thinkingBudget: 0,
         includeThoughts: false,
+        thinkingLevel: "high",
       })
+      expect(prepared.body.cachedContent).toBe("cachedContents/example")
+      expect(prepared.body.safetySettings).toEqual([
+        { category: "HARM_CATEGORY_HATE_SPEECH", threshold: "BLOCK_ONLY_HIGH" },
+      ])
+      expect(prepared.body.serviceTier).toBe("priority")
       expect(filtered.body.generationConfig?.thinkingConfig).toEqual({ includeThoughts: false })
+      expect(defaulted.body.generationConfig?.thinkingConfig).toEqual({
+        includeThoughts: true,
+        thinkingLevel: "high",
+      })
+      expect(emptySafetySettings.body.safetySettings).toEqual([])
     }),
   )
 

packages/ai/test/provider/google-vertex.test.ts

@@ -181,23 +181,6 @@ describe("Google Vertex providers", () => {
     }),
   )
 
-  it.effect("protects the Vertex Messages API version from body overlays", () =>
-    Effect.gen(function* () {
-      const error = yield* compileRequest(
-        LLM.request({
-          model: GoogleVertexMessages.configure({
-            accessToken: "vertex-token",
-            http: { body: { anthropic_version: "wrong" } },
-            project: "vertex-project",
-          }).model("claude-sonnet-4-6"),
-          prompt: "Say hello.",
-        }),
-      ).pipe(Effect.flip)
-
-      expect(error.message).toContain("http.body cannot overlay protocol-owned field(s): anthropic_version")
-    }),
-  )
-
   it.effect("routes tuned Gemini models through their deployed endpoint", () =>
     Effect.gen(function* () {
       const response = yield* LLMClient.generate(

packages/ai/test/provider/openai-compatible-chat.test.ts

@@ -144,6 +144,20 @@ describe("OpenAI-compatible Chat route", () => {
     }),
   )
 
+  it.effect("configures the max tokens request field", () =>
+    Effect.gen(function* () {
+      const compatible = OpenAICompatibleChat.route
+        .with({ provider: "custom", endpoint: { baseURL: "https://api.custom.test/v1" } })
+        .model({ id: "custom-model", compatibility: { maxTokensField: "max_completion_tokens" } })
+      const prepared = yield* compileRequest(
+        LLM.request({ model: compatible, prompt: "Say hello.", generation: { maxTokens: 20 } }),
+      )
+
+      expect(prepared.body).toMatchObject({ max_completion_tokens: 20 })
+      expect(prepared.body).not.toHaveProperty("max_tokens")
+    }),
+  )
+
   it.effect("matches AI SDK compatible tool request body fixture", () =>
     Effect.gen(function* () {
       const prepared = yield* compileRequest(

packages/ai/test/provider/openai-responses-cache.recorded.test.ts

@@ -20,7 +20,7 @@ const cacheRequest = LLM.request({
   system: LARGE_CACHEABLE_SYSTEM,
   prompt: "Say hi.",
   generation: { maxTokens: 16, temperature: 0 },
-  providerOptions: { openai: { promptCacheKey: "recorded-cache-test" } },
+  cache: { mode: "auto", key: "recorded-cache-test" },
 })
 
 const recorded = recordedTests({

packages/ai/test/provider/openai-responses.test.ts

@@ -680,9 +680,9 @@ describe("OpenAI Responses route", () => {
         LLM.request({
           model: OpenAI.configure({ baseURL: "https://api.openai.test/v1/", apiKey: "test" }).model("gpt-5.2"),
           prompt: "think",
+          cache: { mode: "auto", key: "session_123" },
           providerOptions: {
             openai: {
-              promptCacheKey: "session_123",
               reasoningEffort: "high",
               reasoningSummary: "auto",
               include: ["reasoning.encrypted_content"],
@@ -801,17 +801,16 @@ describe("OpenAI Responses route", () => {
     }),
   )
 
-  it.effect("request OpenAI provider options override route defaults", () =>
+  it.effect("maps the request prompt cache key", () =>
     Effect.gen(function* () {
       const prepared = yield* compileRequest(
         LLM.request({
           model: OpenAI.configure({
             baseURL: "https://api.openai.test/v1/",
             apiKey: "test",
-            providerOptions: { openai: { promptCacheKey: "model_cache" } },
           }).model("gpt-4.1-mini"),
           prompt: "no cache",
-          providerOptions: { openai: { promptCacheKey: "request_cache" } },
+          cache: { mode: "auto", key: "request_cache" },
         }),
       )
 

packages/ai/test/provider/openrouter.test.ts

@@ -1,6 +1,6 @@
 import { describe, expect } from "bun:test"
 import { Effect } from "effect"
-import { LLM, Message } from "../../src"
+import { CacheHint, LLM, Message } from "../../src"
 import { LLMClient } from "../../src/route"
 import { compileRequest } from "../../src/route/client"
 import * as OpenRouter from "../../src/providers/openrouter"
@@ -27,10 +27,131 @@ describe("OpenRouter", () => {
         model: "openai/gpt-4o-mini",
         messages: [{ role: "user", content: "Say hello." }],
         stream: true,
+        usage: { include: true },
       })
     }),
   )
 
+  it.effect("lowers the native cache policy to OpenRouter cache controls", () =>
+    Effect.gen(function* () {
+      const prepared = yield* compileRequest(
+        LLM.request({
+          model: OpenRouter.configure({ apiKey: "test-key" }).model("anthropic/claude-sonnet-4.6"),
+          system: [
+            { type: "text", text: "Base agent", cache: new CacheHint({ type: "ephemeral", ttlSeconds: 3_600 }) },
+            { type: "text", text: "Project instructions" },
+          ],
+          tools: [{ name: "lookup", description: "Lookup", inputSchema: { type: "object", properties: {} } }],
+          prompt: "Hello",
+          cache: { mode: "explicit", tools: true, system: true, messages: { tail: 1 } },
+        }),
+      )
+
+      expect(prepared.body).toMatchObject({
+        tools: [{ cache_control: { type: "ephemeral" } }],
+        messages: [
+          {
+            role: "system",
+            content: [
+              { text: "Base agent", cache_control: { type: "ephemeral", ttl: "1h" } },
+              { text: "Project instructions", cache_control: { type: "ephemeral" } },
+            ],
+          },
+          {
+            role: "user",
+            content: [{ text: "Hello", cache_control: { type: "ephemeral" } }],
+          },
+        ],
+      })
+    }),
+  )
+
+  it.effect("lowers manual assistant and tool-result cache hints", () =>
+    Effect.gen(function* () {
+      const prepared = yield* compileRequest(
+        LLM.request({
+          model: OpenRouter.configure({ apiKey: "test-key" }).model("anthropic/claude-sonnet-4.6"),
+          cache: "none",
+          messages: [
+            Message.user("Call the tool"),
+            Message.assistant([
+              { type: "text", text: "Calling", cache: new CacheHint({ type: "ephemeral" }) },
+              { type: "tool-call", id: "call_1", name: "lookup", input: {} },
+            ]),
+            Message.tool({
+              id: "call_1",
+              name: "lookup",
+              result: "Done",
+              cache: new CacheHint({ type: "ephemeral", ttlSeconds: 3_600 }),
+            }),
+          ],
+        }),
+      )
+
+      expect(prepared.body.messages).toMatchObject([
+        { role: "user", content: "Call the tool" },
+        { role: "assistant", content: "Calling", cache_control: { type: "ephemeral" } },
+        { role: "tool", content: '"Done"', cache_control: { type: "ephemeral", ttl: "1h" } },
+      ])
+    }),
+  )
+
+  it.effect("caps manual cache controls at four breakpoints", () =>
+    Effect.gen(function* () {
+      const cache = new CacheHint({ type: "ephemeral" })
+      const prepared = yield* compileRequest(
+        LLM.request({
+          model: OpenRouter.configure({ apiKey: "test-key" }).model("anthropic/claude-sonnet-4.6"),
+          cache: "none",
+          system: [1, 2, 3, 4, 5].map((index) => ({ type: "text" as const, text: `System ${index}`, cache })),
+          prompt: "Hello",
+        }),
+      )
+
+      const system = prepared.body.messages[0]
+      expect(system?.role).toBe("system")
+      expect(
+        system && Array.isArray(system.content)
+          ? system.content.filter((part) => "cache_control" in part && part.cache_control !== undefined)
+          : [],
+      ).toHaveLength(4)
+    }),
+  )
+
+  it.effect("preserves cache policy hints on reasoning-only assistant messages", () =>
+    Effect.gen(function* () {
+      const prepared = yield* compileRequest(
+        LLM.request({
+          model: OpenRouter.configure({ apiKey: "test-key" }).model("anthropic/claude-sonnet-4.6"),
+          cache: { mode: "explicit", messages: "latest-assistant" },
+          messages: [Message.user("Think"), Message.assistant([{ type: "reasoning", text: "Reasoning" }])],
+        }),
+      )
+
+      expect(prepared.body.messages).toMatchObject([
+        { role: "user", content: "Think" },
+        { role: "assistant", cache_control: { type: "ephemeral" } },
+      ])
+    }),
+  )
+
+  it.effect("allows usage accounting to be disabled explicitly", () =>
+    Effect.gen(function* () {
+      const prepared = yield* compileRequest(
+        LLM.request({
+          model: OpenRouter.configure({
+            apiKey: "test-key",
+            providerOptions: { openrouter: { usage: false } },
+          }).model("openai/gpt-4o-mini"),
+          cache: "none",
+          prompt: "Hello",
+        }),
+      )
+
+      expect(prepared.body.usage).toEqual({ include: false })
+    }),
+  )
+
   it.effect("applies OpenRouter payload options from the model helper", () =>
     Effect.gen(function* () {
       const prepared = yield* compileRequest(
@@ -41,11 +162,18 @@ describe("OpenRouter", () => {
               openrouter: {
                 usage: true,
                 reasoning: { effort: "high" },
-                promptCacheKey: "session_123",
+                models: ["anthropic/claude-sonnet-4.6", "google/gemini-3.1-pro"],
+                provider: { order: ["anthropic", "google"], require_parameters: true },
+                plugins: [{ id: "response-healing" }],
+                web_search_options: { engine: "native", max_results: 3 },
+                debug: { echo_upstream_body: true },
+                user: "user_123",
+                future_option: { enabled: true },
               },
             },
           }).model("anthropic/claude-3.7-sonnet:thinking"),
           prompt: "Think briefly.",
+          cache: { mode: "auto", key: "session_123" },
         }),
       )
 
@@ -53,10 +181,54 @@ describe("OpenRouter", () => {
         usage: { include: true },
         reasoning: { effort: "high" },
         prompt_cache_key: "session_123",
+        models: ["anthropic/claude-sonnet-4.6", "google/gemini-3.1-pro"],
+        provider: { order: ["anthropic", "google"], require_parameters: true },
+        plugins: [{ id: "response-healing" }],
+        web_search_options: { engine: "native", max_results: 3 },
+        debug: { echo_upstream_body: true },
+        user: "user_123",
+        future_option: { enabled: true },
       })
     }),
   )
 
+  it.effect("filters invalid known OpenRouter options while preserving extensions", () =>
+    Effect.gen(function* () {
+      const invalid: Record<string, unknown> = {
+        usage: "yes",
+        models: "anthropic/claude-sonnet-4.6",
+        provider: [],
+        plugins: {},
+        web_search_options: [],
+        debug: [],
+        user: 123,
+        reasoning: [],
+        promptCacheKey: 123,
+        future_option: { enabled: true },
+      }
+      const prepared = yield* compileRequest(
+        LLM.request({
+          model: OpenRouter.configure({
+            apiKey: "test-key",
+            providerOptions: { openrouter: invalid },
+          }).model("openai/gpt-4o-mini"),
+          prompt: "Hello",
+        }),
+      )
+
+      expect(prepared.body).toMatchObject({ future_option: { enabled: true } })
+      expect(prepared.body).not.toHaveProperty("usage")
+      expect(prepared.body).not.toHaveProperty("models")
+      expect(prepared.body).not.toHaveProperty("provider")
+      expect(prepared.body).not.toHaveProperty("plugins")
+      expect(prepared.body).not.toHaveProperty("web_search_options")
+      expect(prepared.body).not.toHaveProperty("debug")
+      expect(prepared.body).not.toHaveProperty("user")
+      expect(prepared.body).not.toHaveProperty("reasoning")
+      expect(prepared.body).not.toHaveProperty("prompt_cache_key")
+    }),
+  )
+
   it.effect("preserves the upstream provider finish reason", () =>
     Effect.gen(function* () {
       const model = OpenRouter.configure({ apiKey: "test-key" }).model("anthropic/claude-sonnet-4.6")
@@ -104,6 +276,7 @@ describe("OpenRouter", () => {
       const prepared = yield* compileRequest(
         LLM.request({
           model: OpenRouter.configure({ apiKey: "test-key" }).model("anthropic/claude-sonnet-4.6"),
+          cache: "none",
           messages: [
             Message.assistant([
               {
@@ -137,6 +310,7 @@ describe("OpenRouter", () => {
       const prepared = yield* compileRequest(
         LLM.request({
           model: OpenRouter.configure({ apiKey: "test-key" }).model("anthropic/claude-sonnet-4.6"),
+          cache: "none",
           messages: [
             Message.assistant({
               type: "reasoning",
@@ -162,6 +336,7 @@ describe("OpenRouter", () => {
       const prepared = yield* compileRequest(
         LLM.request({
           model: OpenRouter.configure({ apiKey: "test-key" }).model("anthropic/claude-sonnet-4.6"),
+          cache: "none",
           messages: [
             Message.assistant({
               type: "reasoning",
@@ -183,6 +358,7 @@ describe("OpenRouter", () => {
       const prepared = yield* compileRequest(
         LLM.request({
           model: OpenRouter.configure({ apiKey: "test-key" }).model("anthropic/claude-sonnet-4.6"),
+          cache: "none",
           messages: [Message.assistant({ type: "reasoning", text: "Thinking" })],
         }),
       )

packages/app/package.json

@@ -59,6 +59,7 @@
     "@opencode-ai/sdk": "file:vendor/opencode-ai-sdk-1.18.8-dev.tgz",
     "@opencode-ai/session-ui": "workspace:*",
     "@opencode-ai/ui": "workspace:*",
+    "@opencode-ai/util": "workspace:*",
     "@pierre/trees": "1.0.0-beta.4",
     "@sentry/solid": "catalog:",
     "@shikijs/transformers": "3.9.2",

packages/app/src/components/titlebar-tab-nav.tsx

@@ -4,6 +4,7 @@ import { createResizeObserver } from "@solid-primitives/resize-observer"
 import { createMutation } from "@tanstack/solid-query"
 import { IconButtonV2 } from "@opencode-ai/ui/v2/icon-button-v2"
 import { Icon as IconV2 } from "@opencode-ai/ui/v2/icon"
+import { displayLabel } from "@opencode-ai/util/session-title-fallback"
 import { useGlobal } from "@/context/global"
 import { ServerConnection, serverName } from "@/context/server"
 import { displayName, projectForSession } from "@/pages/layout/helpers"
@@ -54,7 +55,10 @@ export function TabNavItem(props: {
     if (!session) return
     return projectForSession(session, serverCtx()?.projects.list() ?? [])
   })
-  const title = createMemo(() => props.session()?.title ?? props.fallbackTitle)
+  const title = createMemo(() => {
+    const session = props.session()
+    return session ? displayLabel(session) : props.fallbackTitle
+  })
 
   const projectName = createMemo(() => {
     const session = props.session()
@@ -143,7 +147,7 @@ export function TabNavItem(props: {
     if (!canOpenTabRename(props.dragging, editing(), rename.isPending)) return
     const session = props.session()
     if (!session) return
-    titleEl.textContent = session.title
+    titleEl.textContent = session.title ?? ""
     setEditing(true)
 
     requestAnimationFrame(() => {
@@ -302,7 +306,7 @@ export function TabNavItem(props: {
       }}
       data={{
         projectName: projectName(),
-        title: props.session()?.title,
+        title: title(),
         path: previewPath(),
         serverName: serverLabel(),
       }}

packages/app/src/pages/home/home-session-search-controller.ts

@@ -3,6 +3,7 @@ import { useLanguage } from "@/context/language"
 import { serverName } from "@/context/server"
 import { displayName } from "@/pages/layout/helpers"
 import { makeEventListener } from "@solid-primitives/event-listener"
+import { displayLabel } from "@opencode-ai/util/session-title-fallback"
 import { createMemo, onCleanup } from "solid-js"
 import { createStore } from "solid-js/store"
 import type { HomeController } from "./home-controller"
@@ -23,7 +24,7 @@ export function createHomeSessionSearchController(home: HomeController, sessions
     if (!value) return []
     return sessions.data
       .searchRecords()
-      .filter((record) => `${record.session.title} ${record.projectName}`.toLowerCase().includes(value))
+      .filter((record) => `${displayLabel(record.session)} ${record.projectName}`.toLowerCase().includes(value))
   })
   const active = createMemo(() => {
     const records = results()

packages/app/src/pages/home/home-sessions-view.tsx

@@ -6,10 +6,10 @@ import { ButtonV2 } from "@opencode-ai/ui/v2/button-v2"
 import { Icon as IconV2 } from "@opencode-ai/ui/v2/icon"
 import { IconButtonV2 } from "@opencode-ai/ui/v2/icon-button-v2"
 import { TooltipV2 } from "@opencode-ai/ui/v2/tooltip-v2"
+import { displayLabel } from "@opencode-ai/util/session-title-fallback"
 import { useLanguage } from "@/context/language"
 import { ServerConnection } from "@/context/server"
 import { SessionTabAvatarView } from "@/pages/layout/session-tab-avatar"
-import { sessionTitle } from "@/utils/session-title"
 import { shouldOpenSessionInBackground } from "../home-session-open"
 import {
   HomeSessionStatusController,
@@ -344,7 +344,7 @@ function HomeSessionSearchResultRow(
     selected: boolean
   },
 ) {
-  const title = createMemo(() => sessionTitle(props.record.session.title) || props.record.session.id)
+  const title = createMemo(() => displayLabel(props.record.session))
   const showProjectName = () => props.showProjectName() && props.record.projectName
   const key = () => homeSessionSearchKey(props.record)
 
@@ -415,7 +415,7 @@ function HomeSessionGroupHeader(props: {
 }
 
 function HomeSessionRow(props: HomeSessionsViewProps & { record: HomeSessionRecord }) {
-  const title = createMemo(() => sessionTitle(props.record.session.title) || props.record.session.id)
+  const title = createMemo(() => displayLabel(props.record.session))
   const showProjectName = () => props.showProjectName() && props.record.projectName
 
   return (

packages/app/src/pages/layout/sidebar-items.tsx

@@ -5,6 +5,7 @@ import { Icon as IconV2 } from "@opencode-ai/ui/v2/icon"
 import { IconButton } from "@opencode-ai/ui/icon-button"
 import { Spinner } from "@opencode-ai/ui/spinner"
 import { Tooltip } from "@opencode-ai/ui/tooltip"
+import { displayLabel } from "@opencode-ai/util/session-title-fallback"
 import { getFilename } from "@opencode-ai/core/util/path"
 import { A, useParams } from "@solidjs/router"
 import { type Accessor, createMemo, For, type JSX, Match, Show, Switch } from "solid-js"
@@ -14,7 +15,6 @@ import { getAvatarColors, type LocalProject, useLayout } from "@/context/layout"
 import { useNotification } from "@/context/notification"
 import { usePermission } from "@/context/permission"
 import { messageAgentColor } from "@/utils/agent"
-import { sessionTitle } from "@/utils/session-title"
 import { sessionPermissionRequest } from "../session/composer/session-request-tree"
 import { childSessionOnPath, getProjectAvatarSource, hasProjectPermissions } from "./helpers"
 
@@ -104,7 +104,7 @@ const SessionRow = (props: {
   warmPress: () => void
   warmFocus: () => void
 }): JSX.Element => {
-  const title = () => sessionTitle(props.session.title)
+  const title = () => displayLabel(props.session)
 
   return (
     <A
@@ -229,7 +229,7 @@ export const SessionItem = (props: SessionItemProps): JSX.Element => {
               fallback={
                 <Tooltip
                   placement={props.mobile ? "bottom" : "right"}
-                  value={sessionTitle(props.session.title)}
+                  value={displayLabel(props.session)}
                   gutter={10}
                   class="min-w-0 w-full"
                 >

packages/app/src/pages/session/timeline/message-timeline.tsx

@@ -70,7 +70,7 @@ import { legacySessionHref, requireServerKey, sessionHref } from "@/utils/sessio
 import { useSDK } from "@/context/sdk"
 import { useSync } from "@/context/sync"
 import { notifySessionTabsRemoved } from "@/components/titlebar-session-events"
-import { sessionTitle } from "@/utils/session-title"
+import { displayLabel } from "@opencode-ai/util/session-title-fallback"
 import { scheduleConnectedMeasure } from "./measure"
 import { observeElementOffsetReconnectAware } from "./observe-element-offset"
 import { createTimelineProjection } from "./projection"
@@ -296,8 +296,11 @@ export function MessageTimeline(props: {
     if (!id) return
     return sync().session.get(id)
   })
-  const titleValue = createMemo(() => info()?.title)
-  const titleLabel = createMemo(() => sessionTitle(titleValue()))
+  const titleLabel = createMemo(() => {
+    const session = info()
+    if (!session) return
+    return displayLabel(session)
+  })
   const shareUrl = createMemo(() => info()?.share?.url)
   const shareEnabled = createMemo(() => sync().data.config.share !== "disabled")
   const parentID = createMemo(() => info()?.parentID)
@@ -311,7 +314,10 @@ export function MessageTimeline(props: {
     if (!id) return emptyMessages
     return sync().data.message[id] ?? emptyMessages
   })
-  const parentTitle = createMemo(() => sessionTitle(parent()?.title) ?? language.t("command.session.new"))
+  const parentTitle = createMemo(() => {
+    const session = parent()
+    return session ? displayLabel(session) : language.t("command.session.new")
+  })
   const getMsgParts = (msgId: string) => sync().data.part[msgId] ?? emptyParts
   const getMsgPart = (messageID: string, partID: string) => getMsgParts(messageID).find((part) => part.id === partID)
   const childTaskDescription = createMemo(() => {
@@ -329,7 +335,7 @@ export function MessageTimeline(props: {
     if (value) return value
     return language.t("command.session.new")
   })
-  const showHeader = createMemo(() => !!(titleValue() || parentID()))
+  const showHeader = createMemo(() => !!(titleLabel() || parentID()))
   const projection = createTimelineProjection({
     messages: sessionMessages,
     userMessages: () => props.userMessages,
@@ -912,9 +918,10 @@ export function MessageTimeline(props: {
   }
 
   function DialogDeleteSession(props: { sessionID: string }) {
-    const name = createMemo(
-      () => sessionTitle(sync().session.get(props.sessionID)?.title) ?? language.t("command.session.new"),
-    )
+    const name = createMemo(() => {
+      const session = sync().session.get(props.sessionID)
+      return session ? displayLabel(session) : language.t("command.session.new")
+    })
     const handleDelete = async () => {
       await deleteSession(props.sessionID)
       dialog.close()

packages/app/src/pages/session/timeline/observe-element-offset.test.ts

@@ -38,17 +38,19 @@ test("reports a divergent native offset once and ignores equal offsets and unrel
     instance.scrollOffset = offset
   })
 
-  document.body.append(unrelated)
-  unrelated.remove()
-  await frames(2)
-  expect(calls).toEqual([])
-
   route.remove()
   document.body.append(route)
   await new Promise((resolve) => setTimeout(resolve, 0))
   await frames(3)
   expect(calls).toEqual([[0, false]])
 
+  instance.scrollOffset = 79_400
+  document.body.append(unrelated)
+  unrelated.remove()
+  await frames(2)
+  expect(calls).toEqual([[0, false]])
+
+  instance.scrollOffset = 0
   route.remove()
   document.body.append(route)
   await new Promise((resolve) => setTimeout(resolve, 0))

packages/app/src/utils/session-title.ts

@@ -1,7 +0,0 @@
-const pattern = /^(New session|Child session) - \d{4}-\d{2}-\d{2}T\d{2}:\d{2}:\d{2}\.\d{3}Z$/
-
-export function sessionTitle(title?: string) {
-  if (!title) return title
-  const match = title.match(pattern)
-  return match?.[1] ?? title
-}

packages/app/src/utils/session.ts

@@ -1,5 +1,6 @@
 import type { SessionApi, SessionInfo, SessionListInput } from "@opencode-ai/client/promise"
 import type { Session } from "@opencode-ai/sdk/v2/client"
+import { withTimestampedFallback } from "@opencode-ai/util/session-title-fallback"
 
 export function normalizeSessionInfo(input: SessionInfo | Session): Session {
   if (!("location" in input)) return input
@@ -13,7 +14,7 @@ export function normalizeSessionInfo(input: SessionInfo | Session): Session {
     parentID: input.parentID,
     cost: input.cost,
     tokens: input.tokens,
-    title: input.title,
+    title: withTimestampedFallback(input),
     agent: input.agent,
     model: input.model,
     version: "",

packages/cli/script/node-assets.ts

@@ -58,8 +58,9 @@ export async function collectNodeAssets(target: NodeTarget) {
         source: path.join(ptyRoot, relative),
       })),
   ]
-  await Promise.all(assets.map((asset) => stat(asset.source)))
-  return assets
+  const unique = [...new Map(assets.map((asset) => [asset.key, asset])).values()]
+  await Promise.all(unique.map((asset) => stat(asset.source)))
+  return unique
 }
 
 export async function hashNodeAssets(assets: readonly NodeAsset[]) {

packages/cli/src/acp/service.ts

@@ -8,6 +8,7 @@ import {
   type SessionMessageInfo,
   type SkillInfo,
 } from "@opencode-ai/client/promise"
+import { withTimestampedFallback } from "@opencode-ai/util/session-title-fallback"
 import type {
   AgentSideConnection,
   AuthenticateRequest,
@@ -213,7 +214,7 @@ export function make(input: { readonly client: OpenCodeClient; readonly connecti
         sessions: page.data.map((session) => ({
           sessionId: session.id,
           cwd: session.location.directory,
-          title: session.title,
+          title: withTimestampedFallback(session),
           updatedAt: new Date(session.time.updated).toISOString(),
         })),
         ...(page.cursor.next ? { nextCursor: page.cursor.next } : {}),

packages/cli/test/node-assets.test.ts

@@ -0,0 +1,17 @@
+import { expect, test } from "bun:test"
+import { fileURLToPath } from "node:url"
+import { collectNodeAssets } from "../script/node-assets"
+import { nodeTarget, shellParserWasmAssets } from "../src/node/target"
+
+test("collects each SEA asset key once", async () => {
+  const assets = await collectNodeAssets(nodeTarget(process.platform, process.arch))
+  const keys = assets.map((asset) => asset.key)
+
+  expect(new Set(keys).size).toBe(keys.length)
+  expect(assets.filter((asset) => asset.key === shellParserWasmAssets.runtime)).toEqual([
+    {
+      key: shellParserWasmAssets.runtime,
+      source: fileURLToPath(import.meta.resolve(shellParserWasmAssets.runtime)),
+    },
+  ])
+})

packages/client/src/effect/api/api.ts

@@ -118,6 +118,7 @@ export type SessionListOperation<E = never> = (input?: Endpoint5_0Input) => Effe
 
 export type Endpoint5_1Input = {
   readonly id?: Session.ID | undefined
+  readonly title?: string | undefined
   readonly agent?: Agent.ID | undefined
   readonly model?: Model.Ref | undefined
   readonly location?: Location.Ref | undefined

packages/client/src/effect/generated/client.ts

@@ -299,7 +299,13 @@ const Endpoint5_0 = (raw: RawClient["server.session"]) => (input?: Endpoint5_0In
 const Endpoint5_1 = (raw: RawClient["server.session"]) => (input?: Endpoint5_1Input) =>
   preserveEffect<Endpoint5_1Output>()(
     raw["session.create"]({
-      payload: { id: input?.["id"], agent: input?.["agent"], model: input?.["model"], location: input?.["location"] },
+      payload: {
+        id: input?.["id"],
+        title: input?.["title"],
+        agent: input?.["agent"],
+        model: input?.["model"],
+        location: input?.["location"],
+      },
     }).pipe(
       Effect.mapError(mapClientError),
       Effect.map((value) => value.data),

packages/client/src/promise/generated/client.ts

@@ -462,6 +462,7 @@ export function make(options: ClientOptions) {
             path: `/api/session`,
             body: {
               id: input?.["id"],
+              title: input?.["title"],
               agent: input?.["agent"],
               model: input?.["model"],
               location: input?.["location"],

packages/client/src/promise/generated/types.ts

@@ -1757,7 +1757,7 @@ export type SessionInfo = {
   cost: MoneyUSD
   tokens: TokenUsageInfo
   time: { created: number; updated: number; archived?: number }
-  title: string
+  title?: string
   location: LocationRef
   subpath?: string
   revert?: SessionRevert
@@ -2006,7 +2006,7 @@ export type SessionV1Info = {
   cost?: number
   tokens?: { input: number; output: number; reasoning: number; cache: { read: number; write: number } }
   share?: { url: string }
-  title: string
+  title?: string
   agent?: string
   model?: { id: string; providerID: string; variant?: string }
   version: string
@@ -2662,24 +2662,35 @@ export type SessionListOutput = SessionsResponse
 export type SessionCreateInput = {
   readonly id?: {
     readonly id?: string | null
+    readonly title?: string | null
     readonly agent?: string | null
     readonly model?: { readonly id: string; readonly providerID: string; readonly variant?: string } | null
     readonly location?: { readonly directory: string; readonly workspaceID?: string } | null
   }["id"]
+  readonly title?: {
+    readonly id?: string | null
+    readonly title?: string | null
+    readonly agent?: string | null
+    readonly model?: { readonly id: string; readonly providerID: string; readonly variant?: string } | null
+    readonly location?: { readonly directory: string; readonly workspaceID?: string } | null
+  }["title"]
   readonly agent?: {
     readonly id?: string | null
+    readonly title?: string | null
     readonly agent?: string | null
     readonly model?: { readonly id: string; readonly providerID: string; readonly variant?: string } | null
     readonly location?: { readonly directory: string; readonly workspaceID?: string } | null
   }["agent"]
   readonly model?: {
     readonly id?: string | null
+    readonly title?: string | null
     readonly agent?: string | null
     readonly model?: { readonly id: string; readonly providerID: string; readonly variant?: string } | null
     readonly location?: { readonly directory: string; readonly workspaceID?: string } | null
   }["model"]
   readonly location?: {
     readonly id?: string | null
+    readonly title?: string | null
     readonly agent?: string | null
     readonly model?: { readonly id: string; readonly providerID: string; readonly variant?: string } | null
     readonly location?: { readonly directory: string; readonly workspaceID?: string } | null

packages/core/schema.json

@@ -1,9 +1,9 @@
 {
   "version": "7",
   "dialect": "sqlite",
-  "id": "db37a97f-9b5e-4c87-be8b-4feace35136c",
+  "id": "e43ed7e2-b9fc-4178-beae-3646e4a976e1",
   "prevIds": [
-    "a4ba73b4-21bc-41ab-a415-94e2ca38d798"
+    "db37a97f-9b5e-4c87-be8b-4feace35136c"
   ],
   "ddl": [
     {
@@ -1302,7 +1302,7 @@
     },
     {
       "type": "text",
-      "notNull": true,
+      "notNull": false,
       "autoincrement": false,
       "default": null,
       "generated": null,

packages/core/src/aisdk-native.ts

@@ -0,0 +1,131 @@
+export * as AISDKNative from "./aisdk-native"
+
+import { isRecord } from "@opencode-ai/ai/utils/record"
+import { Provider } from "./provider"
+
+export interface Mapping {
+  readonly package: string
+  readonly settings: Readonly<Record<string, unknown>>
+  readonly headers?: Readonly<Record<string, string>>
+  readonly body?: Readonly<Record<string, unknown>>
+}
+
+export function map(packageName: string | undefined, settings: Readonly<Record<string, unknown>>): Mapping | undefined {
+  const baseSettings = mapBaseSettings(settings)
+  switch (packageName) {
+    case "@ai-sdk/google":
+      return {
+        package: "@opencode-ai/ai/providers/google",
+        settings: {
+          ...baseSettings,
+          ...mapAPIKey(settings),
+          ...mapGoogleOptions(settings),
+        },
+      }
+    case "@openrouter/ai-sdk-provider":
+      return mapOpenRouter(settings, baseSettings)
+    case "@ai-sdk/xai":
+      return {
+        package: "@opencode-ai/ai/providers/xai",
+        settings: {
+          ...baseSettings,
+          ...mapAPIKey(settings),
+          ...mapXAIOptions(settings),
+        },
+      }
+  }
+}
+
+function mapBaseSettings(settings: Readonly<Record<string, unknown>>) {
+  return {
+    ...(typeof settings.baseURL === "string" ? { baseURL: settings.baseURL } : {}),
+  }
+}
+
+function mapAPIKey(settings: Readonly<Record<string, unknown>>) {
+  return typeof settings.apiKey === "string" ? { apiKey: settings.apiKey } : {}
+}
+
+function mapGoogleOptions(settings: Readonly<Record<string, unknown>>) {
+  const input = settings.thinkingConfig
+  const thinkingConfig = {
+    ...(isRecord(input) && typeof input.thinkingBudget === "number" ? { thinkingBudget: input.thinkingBudget } : {}),
+    ...(isRecord(input) && typeof input.includeThoughts === "boolean"
+      ? { includeThoughts: input.includeThoughts }
+      : {}),
+    ...(isRecord(input) && typeof input.thinkingLevel === "string" ? { thinkingLevel: input.thinkingLevel } : {}),
+  }
+  const options = {
+    ...(typeof settings.cachedContent === "string" ? { cachedContent: settings.cachedContent } : {}),
+    ...(Array.isArray(settings.safetySettings) ? { safetySettings: settings.safetySettings } : {}),
+    ...(typeof settings.serviceTier === "string" ? { serviceTier: settings.serviceTier } : {}),
+    ...(Object.keys(thinkingConfig).length > 0 ? { thinkingConfig } : {}),
+  }
+  if (Object.keys(options).length === 0) return {}
+  return { providerOptions: { gemini: options } }
+}
+
+function mapOpenRouter(
+  settings: Readonly<Record<string, unknown>>,
+  baseSettings: Readonly<Record<string, unknown>>,
+): Mapping {
+  const headers =
+    Provider.mergeHeaders(
+      {
+        ...(typeof settings.appName === "string" ? { "X-OpenRouter-Title": settings.appName } : {}),
+        ...(typeof settings.appUrl === "string" ? { "HTTP-Referer": settings.appUrl } : {}),
+        ...(isStringRecord(settings.api_keys) && Object.keys(settings.api_keys).length > 0
+          ? { "X-Provider-API-Keys": JSON.stringify(settings.api_keys) }
+          : {}),
+      },
+      isStringRecord(settings.headers) ? settings.headers : undefined,
+    ) ?? {}
+  return {
+    package: "@opencode-ai/ai/providers/openrouter",
+    settings: {
+      ...baseSettings,
+      ...mapAPIKey(settings),
+      ...mapOpenRouterOptions(settings),
+    },
+    ...(Object.keys(headers).length > 0 ? { headers } : {}),
+    ...(isRecord(settings.extraBody) ? { body: settings.extraBody } : {}),
+  }
+}
+
+function mapOpenRouterOptions(settings: Readonly<Record<string, unknown>>) {
+  const options = Object.fromEntries(
+    Object.entries(settings).filter(
+      ([key]) =>
+        ![
+          "apiKey",
+          "api_keys",
+          "appName",
+          "appUrl",
+          "authToken",
+          "baseURL",
+          "chunkTimeout",
+          "compatibility",
+          "extraBody",
+          "fetch",
+          "headers",
+          "promptCacheKey",
+          "timeout",
+        ].includes(key),
+    ),
+  )
+  if (Object.keys(options).length === 0) return {}
+  return { providerOptions: { openrouter: options } }
+}
+
+function isStringRecord(value: unknown): value is Readonly<Record<string, string>> {
+  return isRecord(value) && Object.values(value).every((item) => typeof item === "string")
+}
+
+function mapXAIOptions(settings: Readonly<Record<string, unknown>>) {
+  const options = {
+    ...(typeof settings.reasoningEffort === "string" ? { reasoningEffort: settings.reasoningEffort } : {}),
+    ...(typeof settings.store === "boolean" ? { store: settings.store } : {}),
+  }
+  if (Object.keys(options).length === 0) return {}
+  return { providerOptions: { xai: options } }
+}

packages/core/src/aisdk.ts

@@ -299,8 +299,6 @@ export const locationLayer = Layer.effect(
   }),
 )
 
-export const defaultLayer = locationLayer
-
 function modelFromLanguage(info: Info, language: LanguageModelV3) {
   const packageName = Provider.packageName(info.package!)
   const projected = mapBodyToProviderOptions(info, packageName)
@@ -332,7 +330,7 @@ function modelFromLanguage(info: Info, language: LanguageModelV3) {
               body: projected.body === undefined ? undefined : { ...projected.body },
               headers: info.headers,
             },
-      limits: { context: info.limit.context, output: info.limit.output },
+      limits: { context: info.limit.context, input: info.limit.input, output: info.limit.output },
       providerOptions,
     },
     body: {

packages/core/src/config.ts

@@ -16,7 +16,7 @@ import { Global } from "@opencode-ai/util/global"
 import { Location } from "./location"
 import { AbsolutePath } from "./schema"
 import { ConfigAgent } from "./config/agent"
-import { ConfigAttachments } from "./config/attachments"
+import { ConfigMedia } from "./config/media"
 import { ConfigCompaction } from "./config/compaction"
 import { ConfigCommand } from "./config/command"
 import { ConfigExperimental } from "./config/experimental"
@@ -85,8 +85,8 @@ export class Info extends Schema.Class<Info>("Config.Info")({
   lsp: ConfigLSP.Info.pipe(Schema.optional).annotate({
     description: "Enable built-in language servers or configure server overrides",
   }),
-  attachments: ConfigAttachments.Info.pipe(Schema.optional).annotate({
-    description: "Attachment processing configuration",
+  media: ConfigMedia.Info.pipe(Schema.optional).annotate({
+    description: "Media processing configuration",
   }),
   tool_output: ConfigToolOutput.Info.pipe(Schema.optional).annotate({
     description: "Tool output truncation thresholds",

packages/core/src/config/media.ts

@@ -1,15 +1,15 @@
-export * as ConfigAttachments from "./attachments"
+export * as ConfigMedia from "./media"
 
 import { Schema } from "effect"
 import { PositiveInt } from "../schema"
 
-export class Image extends Schema.Class<Image>("Config.Attachments.Image")({
+export class Image extends Schema.Class<Image>("Config.Media.Image")({
   auto_resize: Schema.Boolean.pipe(Schema.optional),
   max_width: PositiveInt.pipe(Schema.optional),
   max_height: PositiveInt.pipe(Schema.optional),
   max_base64_bytes: PositiveInt.pipe(Schema.optional),
 }) {}
 
-export class Info extends Schema.Class<Info>("Config.Attachments")({
+export class Info extends Schema.Class<Info>("Config.Media")({
   image: Image.pipe(Schema.optional),
 }) {}

packages/core/src/database/migration.gen.ts

@@ -58,5 +58,6 @@ export const migrations = (
     import("./migration/20260722011141_delete_tool_progress_events"),
     import("./migration/20260722170000_canonical_tool_results"),
     import("./migration/20260729022634_session_fork_boundary"),
+    import("./migration/20260730195856_optional_session_title"),
   ])
 ).map((module) => module.default) satisfies DatabaseMigration.Migration[]

packages/core/src/database/migration/20260730195856_optional_session_title.ts

@@ -0,0 +1,14 @@
+import { Effect } from "effect"
+import type { DatabaseMigration } from "../migration"
+
+export default {
+  id: "20260730195856_optional_session_title",
+  up(tx) {
+    return Effect.gen(function* () {
+      yield* tx.run(`ALTER TABLE \`session\` RENAME COLUMN \`title\` TO \`title_old\``)
+      yield* tx.run(`ALTER TABLE \`session\` ADD COLUMN \`title\` text`)
+      yield* tx.run(`UPDATE \`session\` SET \`title\` = \`title_old\``)
+      yield* tx.run(`ALTER TABLE \`session\` DROP COLUMN \`title_old\``)
+    })
+  },
+} satisfies DatabaseMigration.Migration

packages/core/src/database/schema.gen.ts

@@ -217,7 +217,7 @@ export default {
           \`slug\` text NOT NULL,
           \`directory\` text NOT NULL,
           \`path\` text,
-          \`title\` text NOT NULL,
+          \`title\` text,
           \`version\` text NOT NULL,
           \`share_url\` text,
           \`summary_additions\` integer,

packages/core/src/github-copilot/responses/tool/image-generation.ts

@@ -1,4 +1,3 @@
-import { createProviderToolFactoryWithOutputSchema } from "@ai-sdk/provider-utils"
 import { z } from "zod/v4"
 
 export const imageGenerationArgsSchema = z
@@ -24,91 +23,3 @@ export const imageGenerationArgsSchema = z
 export const imageGenerationOutputSchema = z.object({
   result: z.string(),
 })
-
-type ImageGenerationArgs = {
-  /**
-   * Background type for the generated image. Default is 'auto'.
-   */
-  background?: "auto" | "opaque" | "transparent"
-
-  /**
-   * Input fidelity for the generated image. Default is 'low'.
-   */
-  inputFidelity?: "low" | "high"
-
-  /**
-   * Optional mask for inpainting.
-   * Contains image_url (string, optional) and file_id (string, optional).
-   */
-  inputImageMask?: {
-    /**
-     * File ID for the mask image.
-     */
-    fileId?: string
-
-    /**
-     * Base64-encoded mask image.
-     */
-    imageUrl?: string
-  }
-
-  /**
-   * The image generation model to use. Default: gpt-image-1.
-   */
-  model?: string
-
-  /**
-   * Moderation level for the generated image. Default: auto.
-   */
-  moderation?: "auto"
-
-  /**
-   * Compression level for the output image. Default: 100.
-   */
-  outputCompression?: number
-
-  /**
-   * The output format of the generated image. One of png, webp, or jpeg.
-   * Default: png
-   */
-  outputFormat?: "png" | "jpeg" | "webp"
-
-  /**
-   * Number of partial images to generate in streaming mode, from 0 (default value) to 3.
-   */
-  partialImages?: number
-
-  /**
-   * The quality of the generated image.
-   * One of low, medium, high, or auto. Default: auto.
-   */
-  quality?: "auto" | "low" | "medium" | "high"
-
-  /**
-   * The size of the generated image.
-   * One of 1024x1024, 1024x1536, 1536x1024, or auto.
-   * Default: auto.
-   */
-  size?: "auto" | "1024x1024" | "1024x1536" | "1536x1024"
-}
-
-const imageGenerationToolFactory = createProviderToolFactoryWithOutputSchema<
-  {},
-  {
-    /**
-     * The generated image encoded in base64.
-     */
-    result: string
-  },
-  ImageGenerationArgs
->({
-  id: "openai.image_generation",
-  inputSchema: z.object({}),
-  outputSchema: imageGenerationOutputSchema,
-})
-
-export const imageGeneration = (
-  args: ImageGenerationArgs = {}, // default
-) => {
-  return imageGenerationToolFactory(args)
-}

packages/core/src/github-copilot/responses/tool/local-shell.ts

@@ -1,4 +1,3 @@
-import { createProviderToolFactoryWithOutputSchema } from "@ai-sdk/provider-utils"
 import { z } from "zod/v4"
 
 export const localShellInputSchema = z.object({
@@ -15,50 +14,3 @@ export const localShellInputSchema = z.object({
 export const localShellOutputSchema = z.object({
   output: z.string(),
 })
-
-export const localShell = createProviderToolFactoryWithOutputSchema<
-  {
-    /**
-     * Execute a shell command on the server.
-     */
-    action: {
-      type: "exec"
-
-      /**
-       * The command to run.
-       */
-      command: string[]
-
-      /**
-       * Optional timeout in milliseconds for the command.
-       */
-      timeoutMs?: number
-
-      /**
-       * Optional user to run the command as.
-       */
-      user?: string
-
-      /**
-       * Optional working directory to run the command in.
-       */
-      workingDirectory?: string
-
-      /**
-       * Environment variables to set for the command.
-       */
-      env?: Record<string, string>
-    }
-  },
-  {
-    /**
-     * The output of local shell tool call.
-     */
-    output: string
-  },
-  {}
->({
-  id: "openai.local_shell",
-  inputSchema: localShellInputSchema,
-  outputSchema: localShellOutputSchema,
-})

packages/core/src/github-copilot/responses/tool/web-search-preview.ts

@@ -1,4 +1,3 @@
-import { createProviderToolFactory } from "@ai-sdk/provider-utils"
 import { z } from "zod/v4"
 
 // Args validation schema
@@ -39,65 +38,3 @@ export const webSearchPreviewArgsSchema = z.object({
     })
     .optional(),
 })
-
-export const webSearchPreview = createProviderToolFactory<
-  {
-    // Web search doesn't take input parameters - it's controlled by the prompt
-  },
-  {
-    /**
-     * Search context size to use for the web search.
-     * - high: Most comprehensive context, highest cost, slower response
-     * - medium: Balanced context, cost, and latency (default)
-     * - low: Least context, lowest cost, fastest response
-     */
-    searchContextSize?: "low" | "medium" | "high"
-
-    /**
-     * User location information to provide geographically relevant search results.
-     */
-    userLocation?: {
-      /**
-       * Type of location (always 'approximate')
-       */
-      type: "approximate"
-      /**
-       * Two-letter ISO country code (e.g., 'US', 'GB')
-       */
-      country?: string
-      /**
-       * City name (free text, e.g., 'Minneapolis')
-       */
-      city?: string
-      /**
-       * Region name (free text, e.g., 'Minnesota')
-       */
-      region?: string
-      /**
-       * IANA timezone (e.g., 'America/Chicago')
-       */
-      timezone?: string
-    }
-  }
->({
-  id: "openai.web_search_preview",
-  inputSchema: z.object({
-    action: z
-      .discriminatedUnion("type", [
-        z.object({
-          type: z.literal("search"),
-          query: z.string().nullish(),
-        }),
-        z.object({
-          type: z.literal("open_page"),
-          url: z.string(),
-        }),
-        z.object({
-          type: z.literal("find"),
-          url: z.string(),
-          pattern: z.string(),
-        }),
-      ])
-      .nullish(),
-  }),
-})

packages/core/src/github-copilot/responses/tool/web-search.ts

@@ -1,4 +1,3 @@
-import { createProviderToolFactory } from "@ai-sdk/provider-utils"
 import { z } from "zod/v4"
 
 export const webSearchArgsSchema = z.object({
@@ -20,83 +19,3 @@ export const webSearchArgsSchema = z.object({
     })
     .optional(),
 })
-
-export const webSearchToolFactory = createProviderToolFactory<
-  {
-    // Web search doesn't take input parameters - it's controlled by the prompt
-  },
-  {
-    /**
-     * Filters for the search.
-     */
-    filters?: {
-      /**
-       * Allowed domains for the search.
-       * If not provided, all domains are allowed.
-       * Subdomains of the provided domains are allowed as well.
-       */
-      allowedDomains?: string[]
-    }
-
-    /**
-     * Search context size to use for the web search.
-     * - high: Most comprehensive context, highest cost, slower response
-     * - medium: Balanced context, cost, and latency (default)
-     * - low: Least context, lowest cost, fastest response
-     */
-    searchContextSize?: "low" | "medium" | "high"
-
-    /**
-     * User location information to provide geographically relevant search results.
-     */
-    userLocation?: {
-      /**
-       * Type of location (always 'approximate')
-       */
-      type: "approximate"
-      /**
-       * Two-letter ISO country code (e.g., 'US', 'GB')
-       */
-      country?: string
-      /**
-       * City name (free text, e.g., 'Minneapolis')
-       */
-      city?: string
-      /**
-       * Region name (free text, e.g., 'Minnesota')
-       */
-      region?: string
-      /**
-       * IANA timezone (e.g., 'America/Chicago')
-       */
-      timezone?: string
-    }
-  }
->({
-  id: "openai.web_search",
-  inputSchema: z.object({
-    action: z
-      .discriminatedUnion("type", [
-        z.object({
-          type: z.literal("search"),
-          query: z.string().nullish(),
-        }),
-        z.object({
-          type: z.literal("open_page"),
-          url: z.string(),
-        }),
-        z.object({
-          type: z.literal("find"),
-          url: z.string(),
-          pattern: z.string(),
-        }),
-      ])
-      .nullish(),
-  }),
-})
-
-export const webSearch = (
-  args: Parameters<typeof webSearchToolFactory>[0] = {}, // default
-) => {
-  return webSearchToolFactory(args)
-}

packages/core/src/image.ts

@@ -61,7 +61,7 @@ const layer = Layer.effect(
       const image = Object.assign(
         {},
         ...(yield* config.entries()).flatMap((entry) =>
-          entry.type === "document" && entry.info.attachments?.image ? [entry.info.attachments.image] : [],
+          entry.type === "document" && entry.info.media?.image ? [entry.info.media.image] : [],
         ),
       )
       const normalize = yield* loadAdapter

packages/core/src/model-resolver.ts

@@ -12,12 +12,12 @@ import { Auth, type AnyRoute } from "@opencode-ai/ai/route"
 import { Context, Effect, Layer, Schema } from "effect"
 import { produce } from "immer"
 import { AISDK } from "./aisdk"
+import { AISDKNative } from "./aisdk-native"
 import { Catalog } from "./catalog"
 import { Credential } from "./credential"
 import { Integration } from "./integration"
 import { Capabilities, ID, Info, Ref, VariantID } from "./model"
 import { Npm } from "@opencode-ai/util/npm"
-import { OpenAICodex } from "./plugin/provider/openai-codex"
 import { Provider } from "./provider"
 
 export class VariantUnavailableError extends Schema.TaggedErrorClass<VariantUnavailableError>()(
@@ -81,7 +81,7 @@ const withDefaults = (model: Info, route: AnyRoute) =>
     headers: providerHeaders(model),
     providerOptions: providerOptions(model),
     http: model.body === undefined ? undefined : { body: model.body },
-    limits: { context: model.limit.context, output: model.limit.output },
+    limits: { context: model.limit.context, input: model.limit.input, output: model.limit.output },
   })
 
 const providerHeaders = (model: Info) => {
@@ -96,9 +96,7 @@ const providerHeaders = (model: Info) => {
   return Provider.mergeHeaders(generated.size === 0 ? undefined : Object.fromEntries(generated), model.headers)
 }
 
-const providerOptions = (
-  model: Info,
-): { readonly [key: string]: { readonly [key: string]: unknown } } | undefined => {
+const providerOptions = (model: Info): { readonly [key: string]: { readonly [key: string]: unknown } } | undefined => {
   if (!Provider.isAISDK(model.package) || model.settings === undefined) return undefined
   const { apiKey: _, baseURL: _baseURL, ...settings } = model.settings
   if (Object.keys(settings).length === 0) return undefined
@@ -152,12 +150,7 @@ export const fromCatalogModel = (
   const packageName = Provider.packageName(resolved.package)
   const key = apiKey(resolved, credential)
 
-  if (OpenAICodex.isChatGPT(credential) && !Provider.isAISDK(resolved.package) && isNativeOpenAI(resolved.package)) {
-    return Effect.succeed(codexModel(resolved, credential, key))
-  }
-
   if (Provider.isAISDK(resolved.package) && packageName === "@ai-sdk/openai") {
-    if (OpenAICodex.isChatGPT(credential)) return Effect.succeed(codexModel(resolved, credential, key))
     return Effect.succeed(
       withDefaults(resolved, OpenAIResponses.route)
         .with({ auth: key === undefined ? Auth.none : Auth.bearer(key) })
@@ -182,8 +175,10 @@ export const fromCatalogModel = (
         .model({ id: resolved.modelID ?? resolved.id, compatibility: resolved.compatibility }),
     )
   }
-  const native = Provider.isAISDK(resolved.package) ? nativePackage(packageName) : resolved.package
-  if (Provider.isAISDK(resolved.package) && !native) {
+  const configured = { ...resolved.settings, ...credential?.metadata }
+  const mapping = Provider.isAISDK(resolved.package) ? AISDKNative.map(packageName, configured) : undefined
+  const native = mapping?.package ?? resolved.package
+  if (Provider.isAISDK(resolved.package) && !mapping) {
     if (!dependencies?.loadAISDK) return Effect.fail(unsupported(resolved))
     const runtime = produce(resolved, (draft) => {
       draft.settings = Provider.mergeOverlay(draft.settings, {
@@ -201,15 +196,13 @@ export const fromCatalogModel = (
     const module = yield* (dependencies?.loadPackage ?? Provider.loadPackage)(specifier).pipe(
       Effect.mapError(() => unsupported(resolved)),
     )
-    const configured = { ...resolved.settings, ...credential?.metadata }
-    const providerOptions = nativeProviderOptions(packageName, configured)
+    const mapped = mapping?.settings ?? configured
     const settings = {
-      ...(credential ? withoutNativeAuthSettings(configured) : configured),
+      ...(credential ? withoutNativeAuthSettings(mapped) : mapped),
       ...nativeCredentialSettings(specifier, credential),
-      headers: resolved.headers,
-      body: resolved.body,
-      limits: { context: resolved.limit.context, output: resolved.limit.output },
-      ...(providerOptions ? { providerOptions } : {}),
+      headers: Provider.mergeHeaders(mapping?.headers, resolved.headers),
+      body: Provider.mergeOverlay(mapping?.body, resolved.body),
+      limits: { context: resolved.limit.context, input: resolved.limit.input, output: resolved.limit.output },
     }
     return yield* Effect.try({
       try: () => {
@@ -226,30 +219,6 @@ export const fromCatalogModel = (
   })
 }
 
-const nativePackage = (packageName: string | undefined) => {
-  if (packageName === "@ai-sdk/google") return "@opencode-ai/ai/providers/google"
-  if (packageName === "@openrouter/ai-sdk-provider") return "@opencode-ai/ai/providers/openrouter"
-  if (packageName === "@ai-sdk/xai") return "@opencode-ai/ai/providers/xai"
-  return undefined
-}
-
-const nativeProviderOptions = (packageName: string | undefined, settings: Readonly<Record<string, unknown>>) => {
-  const values = Object.fromEntries(
-    Object.entries(settings).filter(
-      ([key]) => !["apiKey", "authToken", "baseURL", "chunkTimeout", "fetch", "timeout"].includes(key),
-    ),
-  )
-  if (Object.keys(values).length === 0) return undefined
-  if (packageName === "@ai-sdk/google") return { gemini: values }
-  if (packageName === "@openrouter/ai-sdk-provider") return { openrouter: values }
-  if (packageName === "@ai-sdk/xai") return { xai: values }
-  return undefined
-}
-
-const isNativeOpenAI = (packageName: string | undefined) =>
-  packageName === "@opencode-ai/ai/providers/openai" ||
-  packageName?.startsWith("@opencode-ai/ai/providers/openai/") === true
-
 const nativeCredentialSettings = (specifier: string, credential: Credential.Value | undefined) => {
   if (!credential) return {}
   if (credential.type === "key") return { apiKey: credential.key }
@@ -271,22 +240,6 @@ const withoutNativeAuthSettings = (settings: Record<string, unknown>) => {
   return rest
 }
 
-const codexModel = (
-  model: Info,
-  credential: Credential.Value | undefined,
-  key: ReturnType<typeof Auth.value> | undefined,
-) => {
-  const account = OpenAICodex.accountID(credential)
-  return withDefaults(model, OpenAIResponses.route)
-    .with({
-      endpoint: { baseURL: OpenAICodex.baseURL },
-      auth: (key === undefined ? Auth.none : Auth.bearer(key)).andThen(
-        account === undefined ? Auth.none : Auth.headers({ "chatgpt-account-id": account }),
-      ),
-    })
-    .model({ id: model.modelID ?? model.id, compatibility: model.compatibility })
-}
-
 const unsupported = (model: Info) =>
   new UnsupportedPackageError({
     providerID: model.providerID,
@@ -311,10 +264,7 @@ export const layer = Layer.effect(
     const integrations = yield* Integration.Service
     const npm = yield* Npm.Service
     const aisdk = yield* AISDK.Service
-    const load = Effect.fn("ModelResolver.resolveModel")(function* (
-      selected: Info,
-      variant?: VariantID,
-    ) {
+    const load = Effect.fn("ModelResolver.resolveModel")(function* (selected: Info, variant?: VariantID) {
       const provider = yield* catalog.provider.get(selected.providerID)
       const connection = yield* integrations.connection.active(
         provider?.integrationID ?? Integration.ID.make(selected.providerID),

packages/core/src/plugin/host.ts

@@ -341,6 +341,7 @@ export const make = Effect.fn("PluginHost.make")(function* (plugin: import("../p
       create: (input) =>
         runtime.session.create({
           id: input?.id,
+          title: input?.title,
           agent: input?.agent,
           model: input?.model,
           location:
@@ -350,8 +351,10 @@ export const make = Effect.fn("PluginHost.make")(function* (plugin: import("../p
       prompt: runtime.session.prompt,
       generate: (input) => runtime.session.generate(input).pipe(Effect.map((text) => ({ text }))),
       command: runtime.session.command,
+      rename: runtime.session.rename,
       synthetic: runtime.session.synthetic,
       interrupt: (input) => runtime.session.interrupt(input.sessionID),
+      wait: (input) => runtime.session.wait(input.sessionID),
     },
   } satisfies Plugin.Context
 })

packages/core/src/plugin/provider/openai-codex.ts

@@ -1,42 +0,0 @@
-export * as OpenAICodex from "./openai-codex"
-
-// TEMPORARY SEAM (#34765): plugins have no hook into LLM route construction, so
-// Codex routing lives in ModelResolver and catalog filtering.
-// in OpenAIPlugin, sharing this module. Once the native provider packages land
-// (#33689/#33925/#34462) this should collapse into the native OpenAI provider.
-// The eligibility rules mirror V1's CodexAuthPlugin allowlist; models.dev has no
-// plan-eligibility data for OpenAI today, but models other vendors' subscriptions
-// as dedicated providers (e.g. zai-coding-plan) - a future openai-chatgpt-plan
-// provider entry could replace the hardcoded rules with catalog data.
-
-/** ChatGPT-plan requests must target the codex backend instead of the public API. */
-export const baseURL = "https://chatgpt.com/backend-api/codex"
-
-const methodIDs: readonly string[] = ["chatgpt-browser", "chatgpt-headless"]
-
-/** Structural credential shape so both core and plugin-facing credential types fit. */
-type CredentialLike = {
-  readonly type: string
-  readonly methodID?: string
-  readonly metadata?: Record<string, unknown> | undefined
-}
-
-export const isChatGPT = (credential: CredentialLike | undefined) =>
-  credential?.type === "oauth" && credential.methodID !== undefined && methodIDs.includes(credential.methodID)
-
-export const accountID = (credential: CredentialLike | undefined) => {
-  if (!isChatGPT(credential)) return undefined
-  const value = credential?.metadata?.accountID
-  return typeof value === "string" ? value : undefined
-}
-
-const allowed = new Set(["gpt-5.5", "gpt-5.3-codex-spark", "gpt-5.4", "gpt-5.4-mini"])
-const disallowed = new Set(["gpt-5.5-pro", "gpt-5.6"])
-
-/** Which API model ids a ChatGPT subscription may call through the codex backend. */
-export const eligible = (apiID: string) => {
-  if (allowed.has(apiID)) return true
-  if (disallowed.has(apiID)) return false
-  const match = apiID.match(/^gpt-(\d+\.\d+)/)
-  return match ? Number.parseFloat(match[1]) > 5.4 : false
-}

packages/core/src/plugin/provider/openai.ts

@@ -10,14 +10,16 @@ import { Model } from "../../model"
 import { OauthCallbackPage } from "../../oauth/page"
 import { Provider } from "../../provider"
 import type { PluginInternal } from "../internal"
-import { OpenAICodex } from "./openai-codex"
 
 const clientID = "app_EMoamEEZ73f0CkXaXp7hrann"
 const issuer = "https://auth.openai.com"
 const callbackPort = 1455
 const pollingSafetyMargin = 3000
+const codexBaseURL = "https://chatgpt.com/backend-api/codex"
 const browserMethodID = Integration.MethodID.make("chatgpt-browser")
 const headlessMethodID = Integration.MethodID.make("chatgpt-headless")
+const codexAllowed = new Set(["gpt-5.5", "gpt-5.3-codex-spark", "gpt-5.4", "gpt-5.4-mini"])
+const codexDisallowed = new Set(["gpt-5.5-pro", "gpt-5.6"])
 
 type Pkce = {
   verifier: string
@@ -164,14 +166,18 @@ export const OpenAIPlugin = define({
   effect: Effect.fn(function* (ctx) {
     const bus = yield* Bus.Service
     const loading = Semaphore.makeUnsafe(1)
-    let chatgpt = false
+    let chatgpt: Credential.OAuth | undefined
 
     const load = Effect.fn("OpenAIPlugin.load")(function* () {
       const connection = yield* ctx.integration.connection.active("openai")
       const credential = connection
         ? yield* ctx.integration.connection.resolve(connection).pipe(Effect.catch(() => Effect.succeed(undefined)))
         : undefined
-      chatgpt = OpenAICodex.isChatGPT(credential)
+      chatgpt =
+        credential?.type === "oauth" &&
+        (credential.methodID === browserMethodID || credential.methodID === headlessMethodID)
+          ? credential
+          : undefined
     })
 
     yield* ctx.integration.transform((draft) => {
@@ -183,16 +189,19 @@ export const OpenAIPlugin = define({
       for (const item of evt.provider.list()) {
         if (!Provider.isAISDK(item.provider.package)) continue
         if (Provider.packageName(item.provider.package) !== "@ai-sdk/openai") continue
-        if (!item.models.has(Model.ID.make("gpt-5-chat-latest"))) continue
-        evt.model.update(item.provider.id, Model.ID.make("gpt-5-chat-latest"), (model) => {
-          // OpenAIPlugin sends OpenAI models through Responses; this alias is a
-          // chat-completions-only model, so hide it only from OpenAI's catalog.
-          model.enabled = false
+        evt.provider.update(item.provider.id, (provider) => {
+          provider.package = "@opencode-ai/ai/providers/openai"
         })
       }
       if (!chatgpt) return
       const item = evt.provider.get(Provider.ID.openai)
       if (!item) return
+      item.provider.settings = Provider.mergeOverlay(item.provider.settings, { baseURL: codexBaseURL })
+      const account = chatgpt.metadata?.accountID
+      item.provider.headers = Provider.mergeHeaders(
+        item.provider.headers,
+        typeof account === "string" ? { "chatgpt-account-id": account } : undefined,
+      )
       for (const model of item.models.values()) {
         // ChatGPT-plan tokens only authorize codex-eligible models, and the
         // subscription covers usage, so hide the rest and zero the cost.
@@ -201,14 +210,32 @@ export const OpenAIPlugin = define({
             draft.enabled = false
             return
           }
-          if (!OpenAICodex.eligible(draft.modelID ?? draft.id)) {
+          const apiID = draft.modelID ?? draft.id
+          const match = apiID.match(/^gpt-(\d+\.\d+)/)
+          if (
+            !codexAllowed.has(apiID) &&
+            (codexDisallowed.has(apiID) || !match || Number.parseFloat(match[1]) <= 5.4)
+          ) {
             draft.enabled = false
             return
           }
           draft.cost = []
+          // Match Codex CLI so context consumption and subscription usage stay consistent between clients.
+          draft.limit = { ...draft.limit, context: 272_000, input: 272_000 }
         })
       }
     })
+    yield* ctx.session.hook("request", (evt) =>
+      Effect.sync(() => {
+        if (!chatgpt || evt.model.providerID !== Provider.ID.openai) return
+        const url = new URL(evt.url)
+        if (url.origin === "https://api.openai.com") {
+          evt.url = `${codexBaseURL}${url.pathname.replace(/^\/v1/, "")}${url.search}`
+        }
+        evt.headers.originator = "opencode"
+        evt.headers["session-id"] = evt.sessionID
+      }),
+    )
 
     const refresh = () => loading.withPermit(load().pipe(Effect.andThen(ctx.catalog.reload())))
     yield* bus.subscribe(Integration.Event.ConnectionUpdated).pipe(
@@ -216,21 +243,6 @@ export const OpenAIPlugin = define({
       Stream.runForEach(refresh),
       Effect.forkScoped({ startImmediately: true }),
     )
-    yield* ctx.aisdk.hook(
-      "sdk",
-      Effect.fn(function* (evt) {
-        if (evt.package !== "@ai-sdk/openai") return
-        const mod = yield* Effect.promise(() => import("@ai-sdk/openai"))
-        evt.sdk = mod.createOpenAI(evt.options)
-      }),
-    )
-    yield* ctx.aisdk.hook(
-      "language",
-      Effect.fn(function* (evt) {
-        if (evt.model.providerID !== Provider.ID.openai) return
-        evt.language = evt.sdk.responses(evt.model.modelID ?? evt.model.id)
-      }),
-    )
   }),
 } satisfies PluginInternal.InternalPlugin)
 

packages/core/src/plugin/runtime.ts

@@ -11,7 +11,17 @@ import { Session } from "../session"
 export interface Interface {
   readonly session: Pick<
     Session.Interface,
-    "get" | "create" | "messages" | "prompt" | "generate" | "command" | "resume" | "interrupt" | "synthetic"
+    | "get"
+    | "create"
+    | "messages"
+    | "prompt"
+    | "generate"
+    | "command"
+    | "rename"
+    | "resume"
+    | "interrupt"
+    | "synthetic"
+    | "wait"
   >
   readonly job: Pick<Job.Interface, "start" | "wait" | "block" | "background" | "cancel">
   readonly location: {
@@ -52,9 +62,11 @@ export const layerWithCell = (cell: Cell) =>
         prompt: (input) => require(cell, (runtime) => runtime.session.prompt(input)),
         generate: (input) => require(cell, (runtime) => runtime.session.generate(input)),
         command: (input) => require(cell, (runtime) => runtime.session.command(input)),
+        rename: (input) => require(cell, (runtime) => runtime.session.rename(input)),
         resume: (sessionID) => require(cell, (runtime) => runtime.session.resume(sessionID)),
         interrupt: (sessionID) => require(cell, (runtime) => runtime.session.interrupt(sessionID)),
         synthetic: (input) => require(cell, (runtime) => runtime.session.synthetic(input)),
+        wait: (sessionID) => require(cell, (runtime) => runtime.session.wait(sessionID)),
       },
       job: {
         start: (input) => require(cell, (runtime) => runtime.job.start(input)),

packages/core/src/session.ts

@@ -374,7 +374,7 @@ const layer = Layer.effect(
           directory: location.directory,
           path: path.relative(project.directory, location.directory).replaceAll("\\", "/"),
           workspaceID: location.workspaceID ? Workspace.ID.make(location.workspaceID) : undefined,
-          title: input.title ?? `New session - ${new Date(now).toISOString()}`,
+          title: input.title,
           agent: input.agent,
           model: input.model
             ? {
@@ -938,7 +938,7 @@ const materializeAttachment = Effect.fn("Session.materializeAttachment")(functio
       : resolved.bytes
   const normalized = yield* normalizeImageAttachment(
     input,
-    Base64.make(Buffer.from(content).toString("base64")),
+    Buffer.from(content).toString("base64"),
     mime,
     image,
   )
@@ -954,11 +954,11 @@ const materializeAttachment = Effect.fn("Session.materializeAttachment")(functio
 
 const normalizeImageAttachment = Effect.fn("Session.normalizeImageAttachment")(function* (
   input: PromptInput.FileAttachment,
-  data: Base64,
+  data: string,
   mime: string,
   image: Effect.Effect<Image.Interface>,
 ) {
-  if (!mime.startsWith("image/")) return { data, mime }
+  if (!mime.startsWith("image/")) return { data: Base64.make(data), mime }
   const service = yield* image
   const label = input.name ?? (input.uri.startsWith("data:") ? "inline attachment" : input.uri)
   const content = { uri: label, content: data, encoding: "base64" as const, mime }

packages/core/src/session/compaction.ts

@@ -10,6 +10,7 @@ import { llmClient } from "../effect/app-node-platform"
 import { SessionEvent } from "./event"
 import type { SessionMessage } from "./message"
 import { SessionModelHeaders } from "./model-headers"
+import { SessionPromptCacheKey } from "./prompt-cache-key"
 import { App } from "../app"
 import { SessionRunnerModel } from "./runner/model"
 import { SessionSchema } from "./schema"
@@ -152,14 +153,11 @@ const settings = (documents: readonly Config.Entry[]) => {
   const configured = documents
     .filter((entry): entry is Config.Document => entry.type === "document")
     .flatMap((entry) => (entry.info.compaction ? [entry.info.compaction] : []))
-  return configured.reduce<Settings>(
-    (result, current) => ({
-      auto: current.auto ?? result.auto,
-      buffer: current.buffer ?? result.buffer,
-      tokens: current.keep?.tokens ?? result.tokens,
-    }),
-    { auto: true, buffer: DEFAULT_BUFFER, tokens: DEFAULT_KEEP_TOKENS },
-  )
+  return {
+    auto: configured.findLast((value) => value.auto !== undefined)?.auto ?? true,
+    buffer: configured.findLast((value) => value.buffer !== undefined)?.buffer ?? DEFAULT_BUFFER,
+    tokens: configured.findLast((value) => value.keep?.tokens !== undefined)?.keep?.tokens ?? DEFAULT_KEEP_TOKENS,
+  }
 }
 
 const select = (
@@ -260,6 +258,7 @@ const make = (dependencies: Dependencies) => {
       .stream(
         LLM.request({
           model: plan.model,
+          cache: { mode: "auto", key: SessionPromptCacheKey.make(plan.session.id) },
           http: { headers: SessionModelHeaders.make(plan.session, dependencies.app) },
           messages: [Message.user(plan.prompt)],
           tools: [],
@@ -350,11 +349,16 @@ const make = (dependencies: Dependencies) => {
         message.type === "assistant" && message.tokens !== undefined,
     )
     if (!last) return false
-    const output = Math.min(input.model.route.defaults.limits?.output ?? 0, OUTPUT_TOKEN_MAX)
+    const limits = input.model.route.defaults.limits
+    const output = Math.min(limits?.output ?? 0, OUTPUT_TOKEN_MAX)
+    const promptCeiling = Math.min(
+      limits?.input === undefined ? Number.POSITIVE_INFINITY : limits.input - config.buffer,
+      context - Math.max(output, config.buffer),
+    )
     const used =
       last.tokens.input + last.tokens.output + last.tokens.reasoning + last.tokens.cache.read + last.tokens.cache.write
     if (used <= 0) return false
-    return used >= context - (output || config.buffer)
+    return used >= promptCeiling
   }
   const compactManual = Effect.fn("SessionCompaction.compactManual")(function* (input: ManualInput) {
     const content = planContent(input.messages, config.tokens)

packages/core/src/session/generate-node.ts

@@ -11,6 +11,7 @@ import { SessionContext } from "./context"
 import { SessionGenerate } from "./generate"
 import { SessionHistory } from "./history"
 import { SessionModelHeaders } from "./model-headers"
+import { SessionPromptCacheKey } from "./prompt-cache-key"
 import { SessionRunnerModel } from "./runner/model"
 import PROMPT_DEFAULT from "./runner/prompt/base.txt"
 import { toLLMMessages } from "./runner/to-llm-message"
@@ -31,9 +32,6 @@ export const layer = Layer.effect(
         const model = yield* models.resolve(selection.session)
         const history = yield* SessionHistory.preview(database.db, selection.session.id, selection.instructions)
         const providerMetadataKey = model.model.route.providerMetadataKey ?? model.model.provider
-        const promptCacheKey = /^ses_[0-9a-f]{64}$/.test(selection.session.id)
-          ? selection.session.id.slice(4)
-          : selection.session.id
         const tools = selection.tools
         const toolDefinitions = tools.definitions
         const toolsByName = new Map(toolDefinitions.map((tool) => [tool.name, tool]))
@@ -71,7 +69,7 @@ export const layer = Layer.effect(
           LLM.request({
             model: model.model,
             http: { headers: SessionModelHeaders.make(selection.session, app) },
-            providerOptions: { openai: { promptCacheKey } },
+            cache: { mode: "auto", key: SessionPromptCacheKey.make(selection.session.id) },
             system: contextEvent.system,
             messages: contextEvent.messages,
             tools: hookedTools,

packages/core/src/session/history.ts

@@ -117,21 +117,20 @@ export const preview = Effect.fn("SessionHistory.preview")(function* (
     .pipe(Effect.catch((error) => (error instanceof Instructions.InitializationBlocked ? error : Effect.die(error))))
 })
 
-/** Returns the session's sole user message, or `undefined` once a second one exists. */
-export const firstUserMessageIfOnly = Effect.fn("SessionHistory.firstUserMessageIfOnly")(function* (
+/** Returns the session's first user message. */
+export const firstUserMessage = Effect.fn("SessionHistory.firstUserMessage")(function* (
   db: DatabaseService,
   sessionID: SessionSchema.ID,
 ) {
-  const rows = yield* db
+  const row = yield* db
     .select()
     .from(SessionMessageTable)
     .where(and(eq(SessionMessageTable.session_id, sessionID), eq(SessionMessageTable.type, "user")))
     .orderBy(asc(SessionMessageTable.seq))
-    .limit(2)
-    .all()
+    .get()
     .pipe(Effect.orDie)
-  if (rows.length !== 1) return undefined
-  const message = yield* decodeMessageRow(rows[0]).pipe(Effect.catch(() => Effect.succeed(undefined)))
+  if (!row) return undefined
+  const message = yield* decodeMessageRow(row).pipe(Effect.catch(() => Effect.succeed(undefined)))
   return message?.type === "user" ? message : undefined
 })
 

packages/core/src/session/info.ts

@@ -17,7 +17,7 @@ export function fromRow(row: typeof SessionTable.$inferSelect): SessionSchema.In
   return SessionSchema.Info.make({
     id: SessionSchema.ID.make(row.id),
     projectID: Project.ID.make(row.project_id),
-    title: row.title,
+    title: row.title ?? undefined,
     parentID: row.parent_id ? SessionSchema.ID.make(row.parent_id) : undefined,
     fork:
       row.fork_session_id && row.fork_boundary

packages/core/src/session/model-request.ts

@@ -1,6 +1,7 @@
 export * as SessionModelRequest from "./model-request"
 
 import { LLM, Message, SystemPart, type LLMRequest } from "@opencode-ai/ai"
+import type { StreamOptions } from "@opencode-ai/ai/route"
 import type { Content } from "@opencode-ai/schema/tool"
 import { SessionError } from "@opencode-ai/schema/session-error"
 import { Cause, Config, Context, Effect, Layer, Result } from "effect"
@@ -13,11 +14,17 @@ import { QuestionTool } from "../tool/plugin/question"
 import { Tool } from "../tool"
 import { SessionContext } from "./context"
 import { SessionModelHeaders } from "./model-headers"
+import { SessionPromptCacheKey } from "./prompt-cache-key"
 import { PromptCacheDiagnostics } from "./prompt-cache-diagnostics"
 import { MAX_STEPS_PROMPT } from "./runner/max-steps"
 import PROMPT_DEFAULT from "./runner/prompt/base.txt"
 import { toLLMMessages } from "./runner/to-llm-message"
 
+const IMAGE_BYTES_TRIGGER = 25 * 1024 * 1024 // 25 MiB
+const IMAGE_BYTES_TARGET = 15 * 1024 * 1024 // 15 MiB
+const IMAGE_REMOVED =
+  "[This image was removed to reduce the request size and is no longer visible. Do not make claims about its contents from memory. If needed, retrieve it again with an available tool or ask the user to attach it again.]"
+
 /** Failures a prepared execution can surface: infrastructure errors plus user declines resurfaced from the defect tunnel. */
 export type ExecuteError = Tool.Error | Permission.DeclinedError | QuestionTool.CancelledError
 
@@ -37,6 +44,7 @@ const declineDefect = (cause: Cause.Cause<Tool.Error>) => {
 
 interface Prepared {
   readonly request: LLMRequest
+  readonly options: StreamOptions
   /**
    * One request-scoped execution operation. Unknown, hook-removed, and
    * step-limit-violating calls fail individually through the same seam.
@@ -92,6 +100,56 @@ export const unsupportedParts = (messages: LLMRequest["messages"], capabilities:
     }),
   )
 
+export const boundImages = (messages: LLMRequest["messages"]) => {
+  const isImage = (mime: string) => mime.toLowerCase().startsWith("image/")
+  const size = (data: string | Uint8Array) =>
+    typeof data === "string" ? Buffer.byteLength(data) : Math.ceil(data.byteLength / 3) * 4
+  const imageBytes = messages.reduce(
+    (total, message) =>
+      total +
+      message.content.reduce((sum, part) => {
+        if (part.type === "media" && isImage(part.mediaType)) return sum + size(part.data)
+        if (part.type !== "tool-result" || part.result.type !== "content") return sum
+        return (
+          sum +
+          part.result.value.reduce(
+            (bytes: number, item: Content) =>
+              bytes + (item.type === "file" && isImage(item.mime) ? Buffer.byteLength(item.uri) : 0),
+            0,
+          )
+        )
+      }, 0),
+    0,
+  )
+  if (imageBytes <= IMAGE_BYTES_TRIGGER) return messages
+
+  let removed = 0
+  return messages.map((message) =>
+    Message.make({
+      ...message,
+      content: message.content.map((part) => {
+        if (part.type === "media" && isImage(part.mediaType) && imageBytes - removed > IMAGE_BYTES_TARGET) {
+          removed += size(part.data)
+          return Message.text(IMAGE_REMOVED)
+        }
+        if (part.type !== "tool-result" || part.result.type !== "content") return part
+        return {
+          ...part,
+          result: {
+            ...part.result,
+            value: part.result.value.map((item: Content) => {
+              if (item.type !== "file" || !isImage(item.mime) || imageBytes - removed <= IMAGE_BYTES_TARGET)
+                return item
+              removed += Buffer.byteLength(item.uri)
+              return { type: "text" as const, text: IMAGE_REMOVED }
+            }),
+          },
+        }
+      }),
+    }),
+  )
+}
+
 /**
  * Builds an outbound model request and captures the tool-call capability that
  * must remain paired with it. It does not execute the request or mutate
@@ -126,7 +184,6 @@ export const layer = Layer.effect(
       // The final Step keeps definitions available to protocols with native "none",
       // preserving their prompt cache prefix. Calls are still rejected at execution.
       const tools = input.context.tools
-      const promptCacheKey = /^ses_[0-9a-f]{64}$/.test(session.id) ? session.id.slice(4) : session.id
       const system = [agent.info.system ? agent.info.system : PROMPT_DEFAULT, input.context.initial]
         .filter((part) => part.length > 0)
         .map(SystemPart.make)
@@ -156,12 +213,32 @@ export const layer = Layer.effect(
         http: {
           headers: SessionModelHeaders.make(session, app),
         },
-        providerOptions: { openai: { promptCacheKey } },
+        cache: { mode: "auto", key: SessionPromptCacheKey.make(session.id) },
         system: contextEvent.system,
-        messages: unsupportedParts(contextEvent.messages, resolved.capabilities),
+        messages: boundImages(unsupportedParts(contextEvent.messages, resolved.capabilities)),
         tools: hookedTools,
         toolChoice: stepLimitReached ? "none" : undefined,
       })
+      const options: StreamOptions = {
+        transform: (request) =>
+          hooks
+            .trigger("session", "request", {
+              sessionID: session.id,
+              agent: agent.id,
+              model: resolved.ref,
+              ...request,
+            })
+            .pipe(
+              Effect.tap((event) =>
+                Effect.sync(() => {
+                  request.url = event.url
+                  request.headers = event.headers
+                  request.body = event.body
+                }),
+              ),
+              Effect.asVoid,
+            ),
+      }
       if (promptCacheSnapshots) {
         const current = PromptCacheDiagnostics.snapshot(request)
         const comparison = PromptCacheDiagnostics.compare(promptCacheSnapshots.get(session.id), current)
@@ -190,6 +267,7 @@ export const layer = Layer.effect(
       }
       return {
         request,
+        options,
         executeTool,
         stepLimitReached,
       }

packages/core/src/session/projector.ts

@@ -43,7 +43,8 @@ type Usage = {
 
 const ForkBatchSize = 500
 
-const forkTitle = (value: string) => {
+const forkTitle = (value?: string) => {
+  if (value === undefined) return
   const match = value.match(/^(.+) \(fork #(\d+)\)$/)
   if (match) return `${match[1]} (fork #${Number.parseInt(match[2], 10) + 1})`
   return `${value} (fork #1)`
@@ -216,7 +217,7 @@ const projectFork = Effect.fn("SessionProjector.projectFork")(function* (
       slug: Slug.create(),
       directory: parent.directory,
       path: parent.path,
-      title: forkTitle(parent.title),
+      title: forkTitle(parent.title ?? undefined),
       agent: parent.agent,
       model: parent.model,
       version: parent.version,

packages/core/src/session/prompt-cache-key.ts

@@ -0,0 +1,6 @@
+export * as SessionPromptCacheKey from "./prompt-cache-key"
+
+import { SessionSchema } from "./schema"
+
+export const make = (sessionID: SessionSchema.ID) =>
+  /^ses_[0-9a-f]{64}$/.test(sessionID) ? sessionID.slice(4) : sessionID

packages/core/src/session/runner/llm.ts

@@ -93,10 +93,9 @@ const layer = Layer.effect(
     const db = (yield* Database.Service).db
     const compaction = yield* SessionCompaction.Service
     const title = yield* SessionTitle.Service
-    // Title generation is a side effect of the first step; it must not delay step continuation.
-    // Tracked per process so repeated wakes before the second user message arrives don't
-    // re-fire a redundant LLM call; `SessionTitle` itself is idempotent based on durable history.
-    const titleStarted = new Set<SessionSchema.ID>()
+    // Title generation is a side effect of a successful step; it must not delay continuation.
+    // The in-flight set coalesces overlapping steps while title presence records success durably.
+    const titlesRunning = new Set<SessionSchema.ID>()
     const forkTitle = yield* FiberSet.makeRuntime<never, void, never>()
     /**
      * Drains eligible manual compaction and user input until the Session becomes idle.
@@ -125,7 +124,7 @@ const layer = Layer.effect(
       let step = 1
       while (true) {
         const result = yield* runStep(sessionID, promotable, step)
-        yield* startTitleOnce(sessionID)
+        if (step === 1) yield* startTitle(sessionID)
         yield* runPendingCompaction(sessionID)
         if (!result.needsContinuation && !(yield* SessionPending.has(db, sessionID, "steer"))) return
         promotable = "steer"
@@ -279,7 +278,7 @@ const layer = Layer.effect(
       // event durably, fork one fiber per local tool call, and hold back a virgin
       // context-overflow provider error so settlement may recover it via compaction.
       let overflowFailure: ProviderErrorEvent | undefined
-      const providerStream = llm.stream(prepared.request).pipe(
+      const providerStream = llm.stream(prepared.request, prepared.options).pipe(
         Stream.runForEach((event) =>
           Effect.gen(function* () {
             if (overflowFailure || publisher.hasProviderError()) return
@@ -481,11 +480,20 @@ const layer = Layer.effect(
       }
     })
 
-    /** Fires title generation once per process after the first step makes a user message visible. */
-    const startTitleOnce = Effect.fnUntraced(function* (sessionID: SessionSchema.ID) {
-      if (titleStarted.has(sessionID)) return
-      titleStarted.add(sessionID)
-      forkTitle(title.generateForFirstPrompt(yield* getSession(sessionID)).pipe(Effect.ignore))
+    /** Starts one title request at a time after a successful step makes user input visible. */
+    const startTitle = Effect.fnUntraced(function* (sessionID: SessionSchema.ID) {
+      if (titlesRunning.has(sessionID)) return
+      titlesRunning.add(sessionID)
+      forkTitle(
+        title.generateForFirstPrompt(sessionID).pipe(
+          Effect.ignore,
+          Effect.ensuring(
+            Effect.sync(() => {
+              titlesRunning.delete(sessionID)
+            }),
+          ),
+        ),
+      )
     })
 
     const getSession = Effect.fn("SessionRunner.getSession")(function* (sessionID: SessionSchema.ID) {

packages/core/src/session/sql.ts

@@ -36,7 +36,7 @@ export const SessionTable = sqliteTable(
     slug: text().notNull(),
     directory: directoryColumn().notNull(),
     path: pathColumn(),
-    title: text().notNull(),
+    title: text(),
     version: text().notNull(),
     share_url: text(),
     summary_additions: integer(),

packages/core/src/session/title.ts

@@ -1,21 +1,25 @@
 export * as SessionTitle from "./title"
 
 import { LLM, LLMClient, LLMError, LLMEvent, Message, type LLMRequest } from "@opencode-ai/ai"
-import { Context, Effect, Layer, Stream } from "effect"
+import { Context, DateTime, Effect, Layer, Stream } from "effect"
 import { Agent } from "../agent"
 import { Database } from "../database/database"
 import { Bus } from "../bus"
 import { makeLocationNode } from "@opencode-ai/util/effect/app-node"
+import { isExactRootFallback } from "@opencode-ai/util/session-title-fallback"
 import { App } from "../app"
 import { llmClient } from "../effect/app-node-platform"
 import { SessionEvent } from "./event"
 import { SessionHistory } from "./history"
 import { SessionModelHeaders } from "./model-headers"
+import { SessionPromptCacheKey } from "./prompt-cache-key"
 import { SessionRunnerModel } from "./runner/model"
 import { SessionSchema } from "./schema"
 import { SessionUsage } from "./usage"
+import { SessionStore } from "./store"
 
 const MAX_LENGTH = 100
+const titleChanged = Symbol("Session title changed")
 
 type Dependencies = {
   readonly app: App.Info
@@ -25,24 +29,33 @@ type Dependencies = {
   }
   readonly agents: Agent.Interface
   readonly models: SessionRunnerModel.Interface
+  readonly store: SessionStore.Interface
 }
 
 export interface Interface {
-  /** Generates a title from the session's first user message and renames the session. Runs at most once per session. */
-  readonly generateForFirstPrompt: (session: SessionSchema.Info) => Effect.Effect<void>
+  /** Generates a title from the session's first user message when the session remains untitled. */
+  readonly generateForFirstPrompt: (sessionID: SessionSchema.ID) => Effect.Effect<void>
 }
 
 export class Service extends Context.Service<Service, Interface>()("@opencode/SessionTitle") {}
 
 const truncate = (value: string) => (value.length <= MAX_LENGTH ? value : `${value.slice(0, MAX_LENGTH - 3)}...`)
+const isUntitled = (session: SessionSchema.Info) =>
+  isExactRootFallback({
+    title: session.title,
+    time: { created: DateTime.toEpochMillis(session.time.created) },
+  })
 
 const make = (dependencies: Dependencies) => {
   const generateForFirstPrompt = Effect.fn("SessionTitle.generateForFirstPrompt")(function* (
     db: Database.Interface["db"],
-    session: SessionSchema.Info,
+    sessionID: SessionSchema.ID,
   ) {
+    const session = yield* dependencies.store.get(sessionID)
+    if (!session) return
     if (session.parentID) return
-    const firstUser = yield* SessionHistory.firstUserMessageIfOnly(db, session.id)
+    if (!isUntitled(session)) return
+    const firstUser = yield* SessionHistory.firstUserMessage(db, session.id)
     if (!firstUser) return
     const agent = yield* dependencies.agents.get(Agent.ID.make("title"))
     if (!agent) return
@@ -68,6 +81,7 @@ const make = (dependencies: Dependencies) => {
       .stream(
         LLM.request({
           model: resolved.model,
+          cache: { mode: "auto", key: SessionPromptCacheKey.make(session.id) },
           http: { headers: SessionModelHeaders.make(session, dependencies.app) },
           system: agent.system,
           messages: [Message.user(firstUser.text)],
@@ -96,10 +110,19 @@ const make = (dependencies: Dependencies) => {
       .map((line) => line.trim())
       .find((line) => line.length > 0)
     if (!title) return
-    yield* dependencies.bus.publish(SessionEvent.Renamed, {
-      sessionID: session.id,
-      title: truncate(title),
-    })
+    const expectedSequence = (yield* Bus.latestSequence(db, sessionID)) + 1
+    const current = yield* dependencies.store.get(sessionID)
+    if (!current || !isUntitled(current)) return
+    yield* dependencies.bus
+      .publish(
+        SessionEvent.Renamed,
+        {
+          sessionID: session.id,
+          title: truncate(title),
+        },
+        { commit: (sequence) => (sequence === expectedSequence ? Effect.void : Effect.die(titleChanged)) },
+      )
+      .pipe(Effect.catchDefect((defect) => (defect === titleChanged ? Effect.void : Effect.die(defect))))
   })
   return { generateForFirstPrompt }
 }
@@ -111,11 +134,12 @@ export const layer = Layer.effect(
     const llm = yield* LLMClient.Service
     const agents = yield* Agent.Service
     const models = yield* SessionRunnerModel.Service
+    const store = yield* SessionStore.Service
     const database = yield* Database.Service
     const app = yield* App.Metadata
-    const title = make({ bus, llm, agents, models, app })
+    const title = make({ bus, llm, agents, models, store, app })
     return Service.of({
-      generateForFirstPrompt: (session) => title.generateForFirstPrompt(database.db, session),
+      generateForFirstPrompt: (sessionID) => title.generateForFirstPrompt(database.db, sessionID),
     })
   }),
 )
@@ -123,5 +147,5 @@ export const layer = Layer.effect(
 export const node = makeLocationNode({
   service: Service,
   layer,
-  deps: [Bus.node, llmClient, Agent.node, SessionRunnerModel.node, Database.node, App.node],
+  deps: [Bus.node, llmClient, Agent.node, SessionRunnerModel.node, SessionStore.node, Database.node, App.node],
 })

packages/core/src/util/module.ts

@@ -1,10 +0,0 @@
-import { createRequire } from "node:module"
-import path from "node:path"
-
-export namespace Module {
-  export function resolve(id: string, dir: string) {
-    try {
-      return createRequire(path.join(dir, "package.json")).resolve(id)
-    } catch {}
-  }
-}

packages/core/src/util/path.ts

@@ -12,12 +12,6 @@ export function getDirectory(path: string | undefined) {
   return parts.slice(0, parts.length - 1).join("/") + "/"
 }
 
-export function getFileExtension(path: string | undefined) {
-  if (!path) return ""
-  const parts = path.split(".")
-  return parts[parts.length - 1]
-}
-
 export function getFilenameTruncated(path: string | undefined, maxLength: number = 20) {
   const filename = getFilename(path)
   if (filename.length <= maxLength) return filename

packages/core/src/v1/config/migrate.ts

@@ -63,7 +63,7 @@ export function migrate(info: typeof ConfigV1.Info.Type) {
     watcher: info.watcher,
     formatter: info.formatter,
     lsp: info.lsp,
-    attachments: info.attachment,
+    media: info.attachment,
     tool_output: info.tool_output,
     mcp: mcp(info),
     compaction: info.compaction && {

packages/core/test/aisdk-native.test.ts

@@ -0,0 +1,138 @@
+import { describe, expect, test } from "bun:test"
+import { AISDKNative } from "@opencode-ai/core/aisdk-native"
+
+describe("AISDKNative", () => {
+  test("maps OpenRouter settings to native destinations", () => {
+    expect(
+      AISDKNative.map("@openrouter/ai-sdk-provider", {
+        appName: "OpenCode",
+        appUrl: "https://opencode.ai",
+        headers: { "x-openrouter-title": "Configured", "x-provider-api-keys": "Configured BYOK" },
+        api_keys: { anthropic: "provider-key" },
+        extraBody: { transforms: ["middle-out"] },
+        models: ["anthropic/claude-sonnet-4.6"],
+        provider: { only: ["anthropic"], require_parameters: true },
+        reasoning: { effort: "high" },
+        future_option: { enabled: true },
+      }),
+    ).toEqual({
+      package: "@opencode-ai/ai/providers/openrouter",
+      settings: {
+        providerOptions: {
+          openrouter: {
+            models: ["anthropic/claude-sonnet-4.6"],
+            provider: { only: ["anthropic"], require_parameters: true },
+            reasoning: { effort: "high" },
+            future_option: { enabled: true },
+          },
+        },
+      },
+      headers: {
+        "x-openrouter-title": "Configured",
+        "HTTP-Referer": "https://opencode.ai",
+        "x-provider-api-keys": "Configured BYOK",
+      },
+      body: { transforms: ["middle-out"] },
+    })
+  })
+
+  test("maps every Google thinking setting", () => {
+    expect(
+      AISDKNative.map("@ai-sdk/google", {
+        cachedContent: "cachedContents/example",
+        safetySettings: [{ category: "HARM_CATEGORY_HARASSMENT", threshold: "BLOCK_NONE" }],
+        serviceTier: "flex",
+        thinkingConfig: {
+          thinkingBudget: 0,
+          includeThoughts: false,
+          thinkingLevel: "high",
+          unknown: true,
+        },
+      }),
+    ).toEqual({
+      package: "@opencode-ai/ai/providers/google",
+      settings: {
+        providerOptions: {
+          gemini: {
+            cachedContent: "cachedContents/example",
+            safetySettings: [{ category: "HARM_CATEGORY_HARASSMENT", threshold: "BLOCK_NONE" }],
+            serviceTier: "flex",
+            thinkingConfig: {
+              thinkingBudget: 0,
+              includeThoughts: false,
+              thinkingLevel: "high",
+            },
+          },
+        },
+      },
+    })
+  })
+
+  test("maps Google thinking settings independently", () => {
+    for (const thinkingConfig of [{ thinkingBudget: -1 }, { includeThoughts: true }, { thinkingLevel: "medium" }]) {
+      expect(AISDKNative.map("@ai-sdk/google", { thinkingConfig })).toMatchObject({
+        settings: { providerOptions: { gemini: { thinkingConfig } } },
+      })
+    }
+  })
+
+  test("maps Google request options without thinking settings", () => {
+    expect(
+      AISDKNative.map("@ai-sdk/google", {
+        cachedContent: "cachedContents/example",
+        safetySettings: [{ category: "HARM_CATEGORY_HARASSMENT", threshold: "BLOCK_NONE" }],
+        serviceTier: "future-tier",
+      }),
+    ).toMatchObject({
+      settings: {
+        providerOptions: {
+          gemini: {
+            cachedContent: "cachedContents/example",
+            safetySettings: [{ category: "HARM_CATEGORY_HARASSMENT", threshold: "BLOCK_NONE" }],
+            serviceTier: "future-tier",
+          },
+        },
+      },
+    })
+  })
+
+  test("maps supported xAI settings", () => {
+    expect(
+      AISDKNative.map("@ai-sdk/xai", {
+        apiKey: "secret",
+        baseURL: "https://xai.example/v1",
+        reasoningEffort: "custom",
+        store: true,
+      }),
+    ).toEqual({
+      package: "@opencode-ai/ai/providers/xai",
+      settings: {
+        apiKey: "secret",
+        baseURL: "https://xai.example/v1",
+        providerOptions: {
+          xai: {
+            reasoningEffort: "custom",
+            store: true,
+          },
+        },
+      },
+    })
+  })
+
+  test("omits invalid and unsupported xAI settings", () => {
+    expect(
+      AISDKNative.map("@ai-sdk/xai", {
+        reasoningEffort: 10,
+        store: "yes",
+        include: ["unknown"],
+        logprobs: true,
+        topLogprobs: 8,
+        previousResponseId: "response-id",
+        searchParameters: { mode: "auto" },
+      }),
+    ).toEqual({
+      package: "@opencode-ai/ai/providers/xai",
+      settings: {},
+    })
+  })
+})

packages/core/test/config/config.test.ts

@@ -803,7 +803,7 @@ describe("Config", () => {
                   custom: { command: ["custom-fmt", "$FILE"], extensions: [".foo"] },
                 },
                 lsp: { typescript: { disabled: true }, custom: { command: ["custom-lsp"], extensions: [".foo"] } },
-                attachments: {
+                media: {
                   image: { auto_resize: false, max_width: 1200, max_height: 900, max_base64_bytes: 1048576 },
                 },
                 tool_output: { max_lines: 1000, max_bytes: 32768 },
@@ -890,7 +890,7 @@ describe("Config", () => {
               typescript: { disabled: true },
               custom: { command: ["custom-lsp"], extensions: [".foo"] },
             })
-            expect(documents[0]?.info.attachments).toEqual({
+            expect(documents[0]?.info.media).toEqual({
               image: { auto_resize: false, max_width: 1200, max_height: 900, max_base64_bytes: 1048576 },
             })
             expect(documents[0]?.info.tool_output).toEqual({ max_lines: 1000, max_bytes: 32768 })
@@ -1097,7 +1097,7 @@ describe("Config", () => {
             expect(documents[0]?.info.references).toEqual({
               docs: { path: "../docs", description: "Use for product documentation", hidden: true },
             })
-            expect(documents[0]?.info.attachments).toEqual({ image: { auto_resize: false, max_width: 1200 } })
+            expect(documents[0]?.info.media).toEqual({ image: { auto_resize: false, max_width: 1200 } })
             expect(documents[0]?.info.providers?.custom).toMatchObject({
               settings: { apiKey: "secret" },
               models: {

packages/core/test/database-migration.test.ts

@@ -26,6 +26,7 @@ import timeSuspendedMigration from "@opencode-ai/core/database/migration/2026070
 import instructionSyncMigration from "@opencode-ai/core/database/migration/20260710025429_instruction_sync"
 import deleteToolProgressEventsMigration from "@opencode-ai/core/database/migration/20260722011141_delete_tool_progress_events"
 import canonicalToolResultsMigration from "@opencode-ai/core/database/migration/20260722170000_canonical_tool_results"
+import optionalSessionTitleMigration from "@opencode-ai/core/database/migration/20260730195856_optional_session_title"
 import { AppNodeBuilder } from "@opencode-ai/core/effect/app-node-builder"
 import { LayerNode } from "@opencode-ai/util/effect/layer-node"
 import { Bus } from "@opencode-ai/core/bus"
@@ -399,6 +400,40 @@ describe("DatabaseMigration", () => {
     ).rejects.toThrow("Database is not empty and has no session table")
   })
 
+  test("makes session titles nullable without deleting dependent rows", async () => {
+    await run(
+      Effect.gen(function* () {
+        const db = yield* makeDb
+        yield* db.run(sql`PRAGMA foreign_keys = ON`)
+        yield* db.run(sql`
+          CREATE TABLE session (
+            id text PRIMARY KEY,
+            title text NOT NULL
+          )
+        `)
+        yield* db.run(sql`
+          CREATE TABLE message (
+            id text PRIMARY KEY,
+            session_id text NOT NULL REFERENCES session(id) ON DELETE CASCADE
+          )
+        `)
+        yield* db.run(sql`INSERT INTO session VALUES ('ses_existing', 'Existing title')`)
+        yield* db.run(sql`INSERT INTO message VALUES ('msg_existing', 'ses_existing')`)
+
+        yield* DatabaseMigration.applyOnly(db, [optionalSessionTitleMigration])
+
+        expect(yield* db.get(sql`SELECT title FROM session WHERE id = 'ses_existing'`)).toEqual({
+          title: "Existing title",
+        })
+        expect(yield* db.get(sql`SELECT id FROM message WHERE id = 'msg_existing'`)).toEqual({ id: "msg_existing" })
+        expect(
+          yield* db.get<{ notnull: number }>(sql`SELECT "notnull" FROM pragma_table_info('session') WHERE name = 'title'`),
+        ).toEqual({ notnull: 0 })
+        expect(yield* db.get<{ foreign_keys: number }>(sql`PRAGMA foreign_keys`)).toEqual({ foreign_keys: 1 })
+      }),
+    )
+  })
+
   test("backfills existing Context Epoch rows to the build agent", async () => {
     await run(
       Effect.gen(function* () {

packages/core/test/model-resolver.test.ts

@@ -1,5 +1,6 @@
 import { describe, expect } from "bun:test"
 import { LLM, Model } from "@opencode-ai/ai"
+import { OpenAIChat } from "@opencode-ai/ai/protocols"
 import { compileRequest } from "@opencode-ai/ai/route/client"
 import { Effect } from "effect"
 import { Headers } from "effect/unstable/http"
@@ -17,6 +18,7 @@ interface ModelOptions {
   readonly headers?: Info["headers"]
   readonly body?: Info["body"]
   readonly variants?: Info["variants"]
+  readonly limit?: Info["limit"]
 }
 
 const model = (packageName: string | undefined, options: ModelOptions = {}) =>
@@ -36,7 +38,7 @@ const model = (packageName: string | undefined, options: ModelOptions = {}) =>
     cost: [],
     status: "active",
     enabled: true,
-    limit: { context: 100, output: 20 },
+    limit: options.limit ?? { context: 100, output: 20 },
   })
 
 describe("ModelResolver", () => {
@@ -44,6 +46,7 @@ describe("ModelResolver", () => {
     Effect.gen(function* () {
       const catalog = model(Provider.aisdk("@ai-sdk/openai"), {
         settings: { baseURL: "https://openai.example/v1" },
+        limit: { context: 100, input: 80, output: 20 },
       })
       const resolved = yield* ModelResolver.fromCatalogModel(catalog)
 
@@ -55,7 +58,7 @@ describe("ModelResolver", () => {
         endpoint: { baseURL: "https://openai.example/v1" },
         defaults: {
           headers: { "x-test": "header" },
-          limits: { context: 100, output: 20 },
+          limits: { context: 100, input: 80, output: 20 },
           http: { body: { custom_extension: { enabled: true } } },
         },
       })
@@ -307,12 +310,12 @@ describe("ModelResolver", () => {
     }),
   )
 
-  it.effect("routes ChatGPT OAuth credentials to the codex backend", () =>
+  it.effect("applies plugin-projected OpenAI endpoint and headers", () =>
     Effect.gen(function* () {
       const resolved = yield* ModelResolver.fromCatalogModel(
-        model(Provider.aisdk("@ai-sdk/openai"), {
-          settings: { baseURL: "https://openai.example/v1" },
-          headers: {},
+        model("@opencode-ai/ai/providers/openai", {
+          settings: { baseURL: "https://chatgpt.com/backend-api/codex" },
+          headers: { "chatgpt-account-id": "acct_123" },
           body: {},
         }),
         Credential.OAuth.make({
@@ -337,37 +340,8 @@ describe("ModelResolver", () => {
         id: "openai-responses",
         endpoint: { baseURL: "https://chatgpt.com/backend-api/codex" },
       })
+      expect(resolved.route.defaults.headers).toMatchObject({ "chatgpt-account-id": "acct_123" })
       expect(headers.authorization).toBe("Bearer chatgpt-token")
-      expect(headers["chatgpt-account-id"]).toBe("acct_123")
-    }),
-  )
-
-  it.effect("routes native OpenAI provider packages with ChatGPT credentials to the codex backend", () =>
-    Effect.gen(function* () {
-      const resolved = yield* ModelResolver.fromCatalogModel(
-        model("@opencode-ai/ai/providers/openai", {
-          settings: { baseURL: "https://openai.example/v1" },
-        }),
-        Credential.OAuth.make({
-          type: "oauth",
-          methodID: Integration.MethodID.make("chatgpt-browser"),
-          access: "chatgpt-token",
-          refresh: "refresh",
-          expires: Date.now() + 60_000,
-          metadata: { accountID: "acct_123" },
-        }),
-      )
-      const headers = yield* resolved.route.auth.apply({
-        request: LLM.request({ model: resolved, prompt: "Hello" }),
-        method: "POST",
-        url: "https://chatgpt.com/backend-api/codex/responses",
-        body: "{}",
-        headers: Headers.empty,
-      })
-
-      expect(resolved.route.endpoint.baseURL).toBe("https://chatgpt.com/backend-api/codex")
-      expect(headers.authorization).toBe("Bearer chatgpt-token")
-      expect(headers["chatgpt-account-id"]).toBe("acct_123")
     }),
   )
 
@@ -407,37 +381,6 @@ describe("ModelResolver", () => {
     }),
   )
 
-  it.effect("routes ChatGPT OAuth credentials without an account id to the codex backend", () =>
-    Effect.gen(function* () {
-      const resolved = yield* ModelResolver.fromCatalogModel(
-        model(Provider.aisdk("@ai-sdk/openai"), {
-          settings: { baseURL: "https://openai.example/v1" },
-          headers: {},
-          body: {},
-        }),
-        Credential.OAuth.make({
-          type: "oauth",
-          methodID: Integration.MethodID.make("chatgpt-headless"),
-          access: "chatgpt-token",
-          refresh: "refresh",
-          expires: Date.now() + 60_000,
-        }),
-      )
-      const request = LLM.request({ model: resolved, prompt: "Hello" })
-      const headers = yield* resolved.route.auth.apply({
-        request,
-        method: "POST",
-        url: "https://chatgpt.com/backend-api/codex/responses",
-        body: "{}",
-        headers: Headers.empty,
-      })
-
-      expect(resolved.route.endpoint.baseURL).toBe("https://chatgpt.com/backend-api/codex")
-      expect(headers.authorization).toBe("Bearer chatgpt-token")
-      expect(headers["chatgpt-account-id"]).toBeUndefined()
-    }),
-  )
-
   it.effect("keeps non-ChatGPT OAuth credentials on the configured endpoint", () =>
     Effect.gen(function* () {
       const resolved = yield* ModelResolver.fromCatalogModel(
@@ -550,16 +493,31 @@ describe("ModelResolver", () => {
     Effect.gen(function* () {
       const native = yield* ModelResolver.fromCatalogModel(model(Provider.aisdk("@ai-sdk/openai")))
       const packages = [
-        ["@ai-sdk/google", "@opencode-ai/ai/providers/google", "gemini"],
-        ["@openrouter/ai-sdk-provider", "@opencode-ai/ai/providers/openrouter", "openrouter"],
-        ["@ai-sdk/xai", "@opencode-ai/ai/providers/xai", "xai"],
+        [
+          "@ai-sdk/google",
+          "@opencode-ai/ai/providers/google",
+          { thinkingConfig: { thinkingLevel: "high" } },
+          { gemini: { thinkingConfig: { thinkingLevel: "high" } } },
+        ],
+        [
+          "@openrouter/ai-sdk-provider",
+          "@opencode-ai/ai/providers/openrouter",
+          { reasoning: { effort: "high" } },
+          { openrouter: { reasoning: { effort: "high" } } },
+        ],
+        [
+          "@ai-sdk/xai",
+          "@opencode-ai/ai/providers/xai",
+          { reasoningEffort: "high" },
+          { xai: { reasoningEffort: "high" } },
+        ],
       ] as const
 
-      yield* Effect.forEach(packages, ([catalogPackage, nativePackage, optionKey]) =>
+      yield* Effect.forEach(packages, ([catalogPackage, nativePackage, sourceOptions, providerOptions]) =>
         ModelResolver.fromCatalogModel(
           model(Provider.aisdk(catalogPackage), {
             modelID: "api-model",
-            settings: { baseURL: "https://provider.example/v1", reasoningEffort: "high" },
+            settings: { baseURL: "https://provider.example/v1", ...sourceOptions },
             headers: { "x-provider": "header" },
             body: { custom: true },
           }),
@@ -576,7 +534,7 @@ describe("ModelResolver", () => {
                     headers: { "x-provider": "header" },
                     body: { custom: true },
                     limits: { context: 100, output: 20 },
-                    providerOptions: { [optionKey]: { reasoningEffort: "high" } },
+                    providerOptions,
                   })
                   return Model.make({ id: modelID, provider: "native-provider", route: native.route })
                 },
@@ -589,6 +547,37 @@ describe("ModelResolver", () => {
     }),
   )
 
+  it.effect("merges mapped OpenRouter headers and body with catalog overlays", () =>
+    ModelResolver.fromCatalogModel(
+      model(Provider.aisdk("@openrouter/ai-sdk-provider"), {
+        settings: {
+          appName: "OpenCode",
+          appUrl: "https://opencode.ai",
+          extraBody: { transforms: ["middle-out"], provider: { sort: "price" } },
+        },
+        headers: { "X-OpenRouter-Title": "Custom" },
+        body: { provider: { only: ["anthropic"] } },
+      }),
+      undefined,
+      {
+        loadPackage: () =>
+          Effect.succeed({
+            model: (modelID, settings) => {
+              expect(settings.headers).toEqual({
+                "HTTP-Referer": "https://opencode.ai",
+                "X-OpenRouter-Title": "Custom",
+              })
+              expect(settings.body).toEqual({
+                transforms: ["middle-out"],
+                provider: { sort: "price", only: ["anthropic"] },
+              })
+              return Model.make({ id: modelID, provider: "openrouter", route: OpenAIChat.route })
+            },
+          }),
+      },
+    ),
+  )
+
   it.effect("loads supported AISDK catalog packages as native routes", () =>
     Effect.gen(function* () {
       const google = yield* ModelResolver.fromCatalogModel(

packages/core/test/plugin/host.ts

@@ -106,8 +106,10 @@ export function host(overrides: Overrides = {}): Plugin.Context {
       prompt: overrides.session?.prompt ?? (() => Effect.die("unused session.prompt")),
       generate: overrides.session?.generate ?? (() => Effect.die("unused session.generate")),
       command: overrides.session?.command ?? (() => Effect.die("unused session.command")),
+      rename: overrides.session?.rename ?? (() => Effect.die("unused session.rename")),
       synthetic: overrides.session?.synthetic ?? (() => Effect.die("unused session.synthetic")),
       interrupt: overrides.session?.interrupt ?? (() => Effect.die("unused session.interrupt")),
+      wait: overrides.session?.wait ?? (() => Effect.die("unused session.wait")),
     },
   }
 }

packages/core/test/plugin/provider-openai.test.ts

@@ -1,7 +1,7 @@
-import { AISDK } from "@opencode-ai/core/aisdk"
 import { Money } from "@opencode-ai/schema/money"
+import { Agent } from "@opencode-ai/schema/agent"
+import { Session } from "@opencode-ai/schema/session"
 import { describe, expect } from "bun:test"
-import type { LanguageModelV3 } from "@ai-sdk/provider"
 import { Effect } from "effect"
 import { Catalog } from "@opencode-ai/core/catalog"
 import { Credential } from "@opencode-ai/core/credential"
@@ -9,6 +9,7 @@ import { Integration } from "@opencode-ai/core/integration"
 import { Model } from "@opencode-ai/core/model"
 import { Plugin } from "@opencode-ai/core/plugin"
 import { PluginHost } from "@opencode-ai/core/plugin/host"
+import { PluginHooks } from "@opencode-ai/core/plugin/hooks"
 import { OpenAIPlugin } from "@opencode-ai/core/plugin/provider/openai"
 import { Provider } from "@opencode-ai/core/provider"
 import { testEffect } from "../lib/effect"
@@ -18,7 +19,6 @@ const it = testEffect(PluginTestLayer)
 
 const addPlugin = Effect.fn(function* () {
   const plugin = yield* Plugin.Service
-  const aisdk = yield* AISDK.Service
   const host = yield* PluginHost.make(plugin)
   const integrations = yield* Integration.Service
   yield* OpenAIPlugin.effect(host).pipe(Effect.provideService(Integration.Service, integrations))
@@ -29,19 +29,6 @@ function required<T>(value: T | undefined): T {
   return value
 }
 
-function fakeSelectorSdk(calls: string[]) {
-  const make = (method: string) => (id: string) => {
-    calls.push(`${method}:${id}`)
-    return { modelId: id, provider: method, specificationVersion: "v3" } as unknown as LanguageModelV3
-  }
-  return {
-    responses: make("responses"),
-    messages: make("messages"),
-    chat: make("chat"),
-    languageModel: make("languageModel"),
-  }
-}
-
 describe("OpenAIPlugin", () => {
   it.effect("registers browser and headless ChatGPT OAuth methods", () =>
     Effect.gen(function* () {
@@ -61,104 +48,6 @@ describe("OpenAIPlugin", () => {
     }),
   )
 
-  it.effect("creates an OpenAI SDK for @ai-sdk/openai using the provider ID as SDK name", () =>
-    Effect.gen(function* () {
-      const plugin = yield* Plugin.Service
-      const aisdk = yield* AISDK.Service
-      yield* addPlugin()
-      const result = yield* aisdk.runSDK({
-        model: Model.Info.make({
-          ...Model.Info.default(Provider.ID.make("custom-openai"), Model.ID.make("gpt-5")),
-          modelID: Model.ID.make("gpt-5"),
-          package: Provider.aisdk("test-provider"),
-        }),
-        package: "@ai-sdk/openai",
-        options: { name: "custom-openai", apiKey: "test" },
-      })
-      expect(result.sdk?.responses("gpt-5").provider).toBe("custom-openai.responses")
-    }),
-  )
-
-  it.effect("ignores non-OpenAI SDK packages", () =>
-    Effect.gen(function* () {
-      const plugin = yield* Plugin.Service
-      const aisdk = yield* AISDK.Service
-      yield* addPlugin()
-      const result = yield* aisdk.runSDK({
-        model: Model.Info.make({
-          ...Model.Info.default(Provider.ID.openai, Model.ID.make("gpt-5")),
-          modelID: Model.ID.make("gpt-5"),
-          package: Provider.aisdk("test-provider"),
-        }),
-        package: "@ai-sdk/openai-compatible",
-        options: { name: "openai" },
-      })
-      expect(result.sdk).toBeUndefined()
-    }),
-  )
-
-  it.effect("uses the Responses API for language models", () =>
-    Effect.gen(function* () {
-      const plugin = yield* Plugin.Service
-      const aisdk = yield* AISDK.Service
-      const calls: string[] = []
-      yield* addPlugin()
-      const result = yield* aisdk.runLanguage({
-        model: Model.Info.make({
-          ...Model.Info.default(Provider.ID.openai, Model.ID.make("alias")),
-          modelID: Model.ID.make("gpt-5"),
-          package: Provider.aisdk("test-provider"),
-        }),
-        sdk: fakeSelectorSdk(calls),
-        options: {},
-      })
-      expect(calls).toEqual(["responses:gpt-5"])
-      expect(result.language).toBeDefined()
-    }),
-  )
-
-  it.effect("ignores non-OpenAI providers", () =>
-    Effect.gen(function* () {
-      const plugin = yield* Plugin.Service
-      const aisdk = yield* AISDK.Service
-      const calls: string[] = []
-      yield* addPlugin()
-      const result = yield* aisdk.runLanguage({
-        model: Model.Info.make({
-          ...Model.Info.default(Provider.ID.anthropic, Model.ID.make("gpt-5")),
-          modelID: Model.ID.make("gpt-5"),
-          package: Provider.aisdk("test-provider"),
-        }),
-        sdk: fakeSelectorSdk(calls),
-        options: {},
-      })
-      expect(calls).toEqual([])
-      expect(result.language).toBeUndefined()
-    }),
-  )
-
-  it.effect("disables gpt-5-chat-latest during catalog transforms", () =>
-    Effect.gen(function* () {
-      const catalog = yield* Catalog.Service
-      yield* catalog.transform((catalog) => {
-        const item = Provider.Info.make({
-          ...Provider.Info.empty(Provider.ID.openai),
-          package: Provider.aisdk("@ai-sdk/openai"),
-        })
-        catalog.provider.update(item.id, (draft) => {
-          draft.package = item.package
-        })
-        catalog.model.update(item.id, Model.ID.make("gpt-5"), () => {})
-        catalog.model.update(item.id, Model.ID.make("gpt-5-chat-latest"), () => {})
-      })
-      yield* addPlugin()
-      expect(required(yield* catalog.model.get(Provider.ID.openai, Model.ID.make("gpt-5"))).enabled).toBe(true)
-      expect(
-        required(yield* catalog.model.get(Provider.ID.openai, Model.ID.make("gpt-5-chat-latest"))).enabled,
-      ).toBe(false)
-    }),
-  )
-
   it.effect("filters the OpenAI catalog to codex-eligible models under a ChatGPT connection", () =>
     Effect.gen(function* () {
       const catalog = yield* Catalog.Service
@@ -172,6 +61,7 @@ describe("OpenAIPlugin", () => {
           draft.package = item.package
         })
         catalog.model.update(item.id, Model.ID.make("gpt-5.5"), (model) => {
+          model.limit = { context: 1_050_000, input: 922_000, output: 128_000 }
           model.cost = [
             {
               input: Money.USDPerMillionTokens.make(1),
@@ -184,12 +74,17 @@ describe("OpenAIPlugin", () => {
           ]
         })
         catalog.model.update(item.id, Model.ID.make("gpt-5.5-pro"), () => {})
+        catalog.model.update(item.id, Model.ID.make("gpt-5.4"), (model) => {
+          model.limit = { context: 1_050_000, input: 922_000, output: 64_000 }
+        })
         catalog.model.update(item.id, Model.ID.make("gpt-5.4-pro"), (model) => {
           model.modelID = Model.ID.make("gpt-5.4")
           model.body = { reasoning: { mode: "pro" } }
         })
         catalog.model.update(item.id, Model.ID.make("gpt-5.6"), () => {})
-        catalog.model.update(item.id, Model.ID.make("gpt-5.6-sol"), () => {})
+        catalog.model.update(item.id, Model.ID.make("gpt-5.6-sol"), (model) => {
+          model.limit = { context: 1_050_000, input: 922_000, output: 128_000 }
+        })
         catalog.model.update(item.id, Model.ID.make("gpt-4.1"), () => {})
       })
       yield* credentials.create({
@@ -205,8 +100,47 @@ describe("OpenAIPlugin", () => {
       })
       yield* addPlugin()
 
+      const request = yield* (yield* PluginHooks.Service).trigger("session", "request", {
+        sessionID: Session.ID.make("ses_test"),
+        agent: Agent.ID.make("build"),
+        model: Model.Ref.make({ providerID: Provider.ID.openai, id: Model.ID.make("gpt-5.5") }),
+        url: "https://api.openai.com/v1/responses",
+        method: "POST",
+        headers: {},
+        body: "{}",
+      })
+      const custom = yield* (yield* PluginHooks.Service).trigger("session", "request", {
+        sessionID: Session.ID.make("ses_test"),
+        agent: Agent.ID.make("build"),
+        model: Model.Ref.make({ providerID: Provider.ID.make("custom-openai"), id: Model.ID.make("gpt-5.5") }),
+        url: "https://custom.example/v1/responses",
+        method: "POST",
+        headers: {},
+        body: "{}",
+      })
+      const proxy = yield* (yield* PluginHooks.Service).trigger("session", "request", {
+        sessionID: Session.ID.make("ses_test"),
+        agent: Agent.ID.make("build"),
+        model: Model.Ref.make({ providerID: Provider.ID.openai, id: Model.ID.make("gpt-5.5") }),
+        url: "https://proxy.example/v1/responses?region=us",
+        method: "POST",
+        headers: {},
+        body: "{}",
+      })
+
+      const provider = required(yield* catalog.provider.get(Provider.ID.openai))
+      expect(provider.package).toBe("@opencode-ai/ai/providers/openai")
+      expect(provider.settings).toMatchObject({ baseURL: "https://chatgpt.com/backend-api/codex" })
+      expect(provider.headers).toMatchObject({ "chatgpt-account-id": "acct_123" })
+      expect(request.url).toBe("https://chatgpt.com/backend-api/codex/responses")
+      expect(request.headers).toMatchObject({ originator: "opencode", "session-id": "ses_test" })
+      expect(custom.headers).toEqual({})
+      expect(proxy.url).toBe("https://proxy.example/v1/responses?region=us")
+      expect(proxy.headers).toMatchObject({ originator: "opencode", "session-id": "ses_test" })
       const eligible = required(yield* catalog.model.get(Provider.ID.openai, Model.ID.make("gpt-5.5")))
+      expect(eligible.package).toBe("@opencode-ai/ai/providers/openai")
       expect(eligible.cost).toEqual([])
+      expect(eligible.limit).toEqual({ context: 272_000, input: 272_000, output: 128_000 })
       expect(eligible.enabled).toBe(true)
       expect(required(yield* catalog.model.get(Provider.ID.openai, Model.ID.make("gpt-5.5-pro"))).enabled).toBe(
         false,
@@ -214,10 +148,15 @@ describe("OpenAIPlugin", () => {
       expect(required(yield* catalog.model.get(Provider.ID.openai, Model.ID.make("gpt-5.4-pro"))).enabled).toBe(
         false,
       )
+      expect(required(yield* catalog.model.get(Provider.ID.openai, Model.ID.make("gpt-5.4"))).limit).toEqual({
+        context: 272_000,
+        input: 272_000,
+        output: 64_000,
+      })
       expect(required(yield* catalog.model.get(Provider.ID.openai, Model.ID.make("gpt-5.6"))).enabled).toBe(false)
-      expect(required(yield* catalog.model.get(Provider.ID.openai, Model.ID.make("gpt-5.6-sol"))).enabled).toBe(
-        true,
-      )
+      const gpt56 = required(yield* catalog.model.get(Provider.ID.openai, Model.ID.make("gpt-5.6-sol")))
+      expect(gpt56.enabled).toBe(true)
+      expect(gpt56.limit).toEqual({ context: 272_000, input: 272_000, output: 128_000 })
       expect(required(yield* catalog.model.get(Provider.ID.openai, Model.ID.make("gpt-4.1"))).enabled).toBe(false)
     }),
   )
@@ -234,7 +173,9 @@ describe("OpenAIPlugin", () => {
         catalog.provider.update(item.id, (draft) => {
           draft.package = item.package
         })
-        catalog.model.update(item.id, Model.ID.make("gpt-5.5"), () => {})
+        catalog.model.update(item.id, Model.ID.make("gpt-5.5"), (model) => {
+          model.limit = { context: 1_050_000, input: 922_000, output: 128_000 }
+        })
         catalog.model.update(item.id, Model.ID.make("gpt-4.1"), () => {})
       })
       yield* credentials.create({
@@ -243,29 +184,23 @@ describe("OpenAIPlugin", () => {
       })
       yield* addPlugin()
 
-      expect(required(yield* catalog.model.get(Provider.ID.openai, Model.ID.make("gpt-5.5"))).enabled).toBe(true)
+      const request = yield* (yield* PluginHooks.Service).trigger("session", "request", {
+        sessionID: Session.ID.make("ses_test"),
+        agent: Agent.ID.make("build"),
+        model: Model.Ref.make({ providerID: Provider.ID.openai, id: Model.ID.make("gpt-5.5") }),
+        url: "https://api.openai.com/v1/responses",
+        method: "POST",
+        headers: {},
+        body: "{}",
+      })
+
+      const model = required(yield* catalog.model.get(Provider.ID.openai, Model.ID.make("gpt-5.5")))
+      expect(model.package).toBe("@opencode-ai/ai/providers/openai")
+      expect(model.enabled).toBe(true)
+      expect(model.limit).toEqual({ context: 1_050_000, input: 922_000, output: 128_000 })
+      expect(request.headers).toEqual({})
       expect(required(yield* catalog.model.get(Provider.ID.openai, Model.ID.make("gpt-4.1"))).enabled).toBe(true)
     }),
   )
 
-  it.effect("does not disable gpt-5-chat-latest for non-OpenAI providers", () =>
-    Effect.gen(function* () {
-      const catalog = yield* Catalog.Service
-      yield* catalog.transform((catalog) => {
-        const item = Provider.Info.make({
-          ...Provider.Info.empty(Provider.ID.make("custom-openai")),
-          package: Provider.aisdk("test-provider"),
-        })
-        catalog.provider.update(item.id, (draft) => {
-          draft.package = item.package
-        })
-        catalog.model.update(item.id, Model.ID.make("gpt-5-chat-latest"), () => {})
-      })
-      yield* addPlugin()
-      expect(
-        required(yield* catalog.model.get(Provider.ID.make("custom-openai"), Model.ID.make("gpt-5-chat-latest")))
-          .enabled,
-      ).toBe(true)
-    }),
-  )
 })

packages/core/test/session-compaction.test.ts

@@ -19,9 +19,11 @@ import { Session } from "@opencode-ai/core/session"
 import { Project } from "@opencode-ai/core/project"
 import { ProjectTable } from "@opencode-ai/core/project/sql"
 import { App } from "@opencode-ai/core/app"
+import { Agent } from "@opencode-ai/core/agent"
+import { Location } from "@opencode-ai/core/location"
 import { AbsolutePath } from "@opencode-ai/core/schema"
 import { Money } from "@opencode-ai/schema/money"
-import { DateTime, Effect, Fiber, Layer, Stream } from "effect"
+import { DateTime, Effect, Fiber, Layer, Schema, Stream } from "effect"
 import { asc, eq } from "drizzle-orm"
 import { testEffect } from "./lib/effect"
 
@@ -130,6 +132,52 @@ test("compaction prompt requires the checkpoint headings in order", () => {
   expect(prompt).toContain("Keep every section, even when empty.")
 })
 
+it.effect("auto compaction reserves a buffer below the prompt ceiling", () =>
+  Effect.gen(function* () {
+    const compaction = yield* SessionCompaction.Service
+    const session = Session.Info.make({
+      id: Session.ID.make("ses_input_limit"),
+      projectID: Project.ID.global,
+      cost: Money.USD.zero,
+      tokens: { input: 0, output: 0, reasoning: 0, cache: { read: 0, write: 0 } },
+      time: { created: DateTime.makeUnsafe(0), updated: DateTime.makeUnsafe(0) },
+      location: Location.Ref.make({ directory: AbsolutePath.make("/tmp") }),
+    })
+    const input = (tokens: number, limits: { context: number; input?: number; output: number }) => ({
+      session,
+      model: Model.make({
+        id: "test-model",
+        provider: "test-provider",
+        route: OpenAIChat.route.with({ limits }),
+      }),
+      cost: [],
+      messages: [
+        Schema.decodeUnknownSync(SessionMessage.Assistant)({
+          id: SessionMessage.ID.make("msg_assistant"),
+          type: "assistant",
+          agent: Agent.defaultID,
+          model: { id: "test-model", providerID: "test-provider" },
+          content: [],
+          tokens: { input: tokens, output: 0, reasoning: 0, cache: { read: 0, write: 0 } },
+          time: { created: 0, completed: 0 },
+        }),
+      ],
+    })
+
+    const inputLimited = { context: 400_000, input: 272_000, output: 128_000 }
+    expect(compaction.required(input(251_999, inputLimited))).toBe(false)
+    expect(compaction.required(input(252_000, inputLimited))).toBe(true)
+
+    const contextLimited = { context: 100_000, output: 10_000 }
+    expect(compaction.required(input(79_999, contextLimited))).toBe(false)
+    expect(compaction.required(input(80_000, contextLimited))).toBe(true)
+
+    const outputLimited = { context: 100_000, output: 30_000 }
+    expect(compaction.required(input(69_999, outputLimited))).toBe(false)
+    expect(compaction.required(input(70_000, outputLimited))).toBe(true)
+  }),
+)
+
 it.effect("manual compaction summarizes short context instead of no-op", () =>
   Effect.gen(function* () {
     requests = []
@@ -187,6 +235,7 @@ it.effect("manual compaction summarizes short context instead of no-op", () =>
     expect(Array.from(yield* Fiber.join(delta)).map((event) => event.data.text)).toEqual(["manual summary"])
 
     expect(requests).toHaveLength(1)
+    expect(requests[0]?.cache).toEqual({ mode: "auto", key: sessionID })
     expect(requests[0]?.http?.headers).toEqual({
       "x-session-affinity": sessionID,
       "X-Session-Id": sessionID,

packages/core/test/session-create.test.ts

@@ -71,6 +71,27 @@ function withTmp<A, E, R>(f: (directory: string) => Effect.Effect<A, E, R>) {
 }
 
 describe("Session.create", () => {
+  it.effect("persists a missing title until one is generated or supplied", () =>
+    Effect.gen(function* () {
+      const session = yield* Session.Service
+      const { db } = yield* Database.Service
+
+      const created = yield* session.create({ location })
+      const row = yield* db.select().from(SessionTable).where(eq(SessionTable.id, created.id)).get().pipe(Effect.orDie)
+      const event = yield* db
+        .select({ data: EventTable.data })
+        .from(EventTable)
+        .where(eq(EventTable.aggregate_id, created.id))
+        .get()
+        .pipe(Effect.orDie)
+
+      expect(created.title).toBeUndefined()
+      expect(row?.title).toBeNull()
+      expect(event?.data).not.toHaveProperty("info.title")
+      expect((yield* session.create({ location, title: "Explicit title" })).title).toBe("Explicit title")
+    }),
+  )
+
   it.effect("creates a fresh projected session when the ID is omitted", () =>
     Effect.gen(function* () {
       const session = yield* Session.Service
@@ -296,6 +317,23 @@ describe("Session.create", () => {
     }),
   )
 
+  it.effect("keeps a fork untitled when its parent is untitled", () =>
+    Effect.gen(function* () {
+      const session = yield* Session.Service
+      const bus = yield* Bus.Service
+      const { db } = yield* Database.Service
+      const parent = yield* session.create({ location })
+      yield* session.prompt({ sessionID: parent.id, text: "First", resume: false })
+      yield* SessionPending.promote(db, bus, parent.id, "steer")
+
+      const forked = yield* session.fork({ sessionID: parent.id, boundary: { type: "through" } })
+      const row = yield* db.select().from(SessionTable).where(eq(SessionTable.id, forked.id)).get().pipe(Effect.orDie)
+
+      expect(forked.title).toBeUndefined()
+      expect(row?.title).toBeNull()
+    }),
+  )
+
   it.effect("rejects forking an empty session", () =>
     Effect.gen(function* () {
       const session = yield* Session.Service

packages/core/test/session-generate.test.ts

@@ -287,7 +287,7 @@ it.effect("generates from fresh settled Session context without durable mutation
     expect(requests[0]?.system[0]?.text).toBe("Hooked system")
     expect(requests[0]?.system.map((part) => part.text)).toContain("Initial context")
     expect(requests[0]?.http?.headers).toMatchObject({ "X-Session-Id": sessionID })
-    expect(requests[0]?.providerOptions).toMatchObject({ openai: { promptCacheKey: sessionID } })
+    expect(requests[0]?.cache).toEqual({ mode: "auto", key: sessionID })
     const instructionUpdates = requests[0]?.messages.flatMap((message) =>
       message.role === "system"
         ? message.content.flatMap((content) => (content.type === "text" ? [content.text] : []))

packages/core/test/session-model-request.test.ts

@@ -1,6 +1,6 @@
 import { describe, expect, test } from "bun:test"
 import { Message, ToolResultPart } from "@opencode-ai/ai"
-import { unsupportedParts } from "@opencode-ai/core/session/model-request"
+import { boundImages, unsupportedParts } from "@opencode-ai/core/session/model-request"
 
 const capabilities = (input: string[]) => ({ tools: true, input, output: ["text"] })
 
@@ -62,3 +62,51 @@ describe("SessionModelRequest.unsupportedParts", () => {
     expect(unsupportedParts([message], capabilities(["text", "image"]))[0]?.content).toEqual(message.content)
   })
 })
+
+describe("SessionModelRequest.boundImages", () => {
+  test("preserves images below the trigger", () => {
+    const messages = [Message.user({ type: "media", mediaType: "image/png", data: "aGVsbG8=" })]
+    expect(boundImages(messages)).toBe(messages)
+  })
+
+  test("replaces oldest images until the retained payload reaches the target", () => {
+    const image = "a".repeat(9 * 1024 * 1024)
+    const messages = [
+      Message.user({ type: "media", mediaType: "image/png", data: image, filename: "first.png" }),
+      Message.user({ type: "media", mediaType: "image/png", data: image, filename: "second.png" }),
+      Message.user({ type: "media", mediaType: "image/png", data: image, filename: "third.png" }),
+    ]
+    const result = boundImages(messages)
+
+    expect(result[0]?.content[0]).toMatchObject({ type: "text" })
+    expect(result[1]?.content[0]).toMatchObject({ type: "text" })
+    expect(result[2]?.content[0]).toMatchObject({ type: "media", filename: "third.png" })
+  })
+
+  test("replaces images nested in tool results", () => {
+    const image = "a".repeat(13 * 1024 * 1024)
+    const result = boundImages([
+      Message.tool(
+        ToolResultPart.make({
+          id: "call_1",
+          name: "read",
+          result: {
+            type: "content",
+            value: [
+              { type: "file", uri: `data:image/png;base64,${image}`, mime: "image/png", name: "first.png" },
+              { type: "file", uri: `data:image/png;base64,${image}`, mime: "image/png", name: "second.png" },
+            ],
+          },
+        }),
+      ),
+    ])
+
+    expect(result[0]?.content[0]).toMatchObject({
+      type: "tool-result",
+      result: {
+        type: "content",
+        value: [{ type: "text" }, { type: "file", name: "second.png" }],
+      },
+    })
+  })
+})

packages/core/test/session-prompt.test.ts

@@ -26,8 +26,8 @@ import { SessionPendingTable, SessionMessageTable, SessionTable } from "@opencod
 import { SessionStore } from "@opencode-ai/core/session/store"
 import { LocationServiceMap } from "@opencode-ai/core/location-service-map"
 import type { LocationServices } from "@opencode-ai/core/location-services"
+import { Image } from "@opencode-ai/core/image"
 import { testEffect } from "./lib/effect"
-import { imagePassthrough } from "./lib/image"
 
 const executionCalls: Session.ID[] = []
 const interruptCalls: Session.ID[] = []
@@ -58,7 +58,10 @@ const locations = Layer.effect(
     () =>
       // Attachment admission only needs the location-scoped Image service.
       // oxlint-disable-next-line typescript-eslint/no-unsafe-type-assertion
-      imagePassthrough as unknown as Layer.Layer<LocationServices>,
+      Layer.mock(Image.Service, {
+        normalize: (_resource, content) =>
+          Effect.succeed(content.content.length > 5 * 1024 * 1024 ? { ...content, content: "AA==" } : content),
+      }) as unknown as Layer.Layer<LocationServices>,
   ),
 )
 const it = testEffect(
@@ -386,6 +389,35 @@ describe("Session.prompt", () => {
     }),
   )
 
+  it.effect("normalizes large image content before validating persisted Base64", () =>
+    Effect.gen(function* () {
+      yield* setup
+      const session = yield* Session.Service
+      const pixel = Buffer.from(
+        "iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAQAAAC1HAwCAAAAC0lEQVR42mNk+A8AAQUBAScY42YAAAAASUVORK5CYII=",
+        "base64",
+      )
+      const bytes = Buffer.concat([pixel, Buffer.alloc(4_323_030 - pixel.length)])
+      const data = bytes.toString("base64")
+      expect(data).toHaveLength(5_764_040)
+
+      const message = yield* session.prompt({
+        sessionID,
+        text: "Inspect this image",
+        files: [{ uri: `data:image/png;base64,${data}` }],
+        resume: false,
+      })
+
+      expect(message.data.files).toEqual([
+        {
+          data: "AA==",
+          mime: "image/png",
+          source: { type: "inline" },
+        },
+      ])
+    }),
+  )
+
   it.effect("sniffs data URL content instead of trusting its declared MIME", () =>
     Effect.gen(function* () {
       yield* setup

packages/core/test/session-runner.test.ts

@@ -796,6 +796,59 @@ const verifyPartialFlushOnInterruption = (kind: FragmentKind) =>
   })
 
 describe("SessionRunnerLLM", () => {
+  it.effect("retries title generation from the first prompt after execution and title failures", () =>
+    Effect.gen(function* () {
+      const session = yield* setup
+      const agents = yield* Agent.Service
+      const { db } = yield* Database.Service
+      yield* db.update(SessionTable).set({ title: null }).where(eq(SessionTable.id, sessionID)).run().pipe(Effect.orDie)
+      yield* agents.transform((draft) =>
+        draft.update(Agent.ID.make("title"), (agent) => {
+          agent.mode = "primary"
+          agent.hidden = true
+          agent.system = "Generate a title."
+        }),
+      )
+
+      yield* admit(session, "First prompt")
+      yield* TestLLM.push(Stream.fail(invalidRequest()))
+      expect((yield* session.resume(sessionID).pipe(Effect.exit))._tag).toBe("Failure")
+
+      yield* admit(session, "Second prompt")
+      const titleFailed = yield* Deferred.make<void>()
+      yield* TestLLM.push(
+        TestLLM.text("Recovered", "text-recovered"),
+        Stream.make(LLMEvent.providerError({ message: "Title provider unavailable" })).pipe(
+          Stream.ensuring(Deferred.succeed(titleFailed, undefined)),
+        ),
+      )
+      yield* session.resume(sessionID)
+      yield* Deferred.await(titleFailed)
+      yield* Effect.yieldNow
+      expect((yield* session.get(sessionID)).title).toBeUndefined()
+
+      const bus = yield* Bus.Service
+      const renamed = yield* bus.subscribe(SessionEvent.Renamed).pipe(
+        Stream.filter((event) => event.data.sessionID === sessionID),
+        Stream.take(1),
+        Stream.runCollect,
+        Effect.forkScoped({ startImmediately: true }),
+      )
+      yield* admit(session, "Third prompt")
+      yield* TestLLM.push(
+        TestLLM.text("Recovered again", "text-recovered-again"),
+        TestLLM.text("Generated title", "text-title"),
+      )
+      yield* session.resume(sessionID)
+      yield* Fiber.join(renamed)
+
+      expect(requests).toHaveLength(5)
+      expect(requests[2]?.messages).toContainEqual(Message.user("First prompt"))
+      expect(requests[4]?.messages).toContainEqual(Message.user("First prompt"))
+      expect((yield* session.get(sessionID)).title).toBe("Generated title")
+    }),
+  )
+
   it.effect("applies session context hooks without exposing unavailable tools", () =>
     Effect.gen(function* () {
       const session = yield* setup
@@ -3157,7 +3210,7 @@ describe("SessionRunnerLLM", () => {
       yield* stream.started
 
       expect(requests).toHaveLength(2)
-      expect(requests.map((request) => request.providerOptions?.openai?.promptCacheKey)).toEqual([
+      expect(requests.map((request) => (request.cache !== "none" ? request.cache?.key : undefined))).toEqual([
         sessionID,
         otherSessionID,
       ])
@@ -3188,7 +3241,7 @@ describe("SessionRunnerLLM", () => {
       yield* session.resume(longSessionID)
       yield* session.resume(otherLongSessionID)
 
-      const keys = requests.map((request) => request.providerOptions?.openai?.promptCacheKey)
+      const keys = requests.map((request) => (request.cache !== "none" ? request.cache?.key : undefined))
       expect(keys).toEqual([longSessionID.slice(4), otherLongSessionID.slice(4)])
       expect(keys.every((key) => typeof key === "string" && key.length === 64)).toBe(true)
       expect(keys[0]).not.toBe(keys[1])

packages/core/test/session-title.test.ts

@@ -20,7 +20,7 @@ import { ProjectTable } from "@opencode-ai/core/project/sql"
 import { App } from "@opencode-ai/core/app"
 import { AbsolutePath } from "@opencode-ai/core/schema"
 import { Money } from "@opencode-ai/schema/money"
-import { Effect, Layer, Stream } from "effect"
+import { Deferred, Effect, Fiber, Layer, Stream } from "effect"
 import { testEffect } from "./lib/effect"
 
 let requests: LLMRequest[] = []
@@ -39,27 +39,30 @@ const cost = [
     },
   },
 ]
+const successfulTitle = () =>
+  Stream.make(
+    LLMEvent.textDelta({ id: "title", text: "Generated Title\n" }),
+    LLMEvent.stepFinish({
+      index: 0,
+      reason: { normalized: "stop" },
+      usage: {
+        inputTokens: 15,
+        outputTokens: 6,
+        nonCachedInputTokens: 10,
+        cacheReadInputTokens: 3,
+        cacheWriteInputTokens: 2,
+        reasoningTokens: 2,
+      },
+    }),
+    LLMEvent.finish({
+      reason: { normalized: "stop" },
+    }),
+  )
+let titleStream: () => Stream.Stream<LLMEvent> = successfulTitle
 const client = Layer.mock(LLMClient.Service)({
   stream: (request: LLMRequest) => {
     requests.push(request)
-    return Stream.make(
-      LLMEvent.textDelta({ id: "title", text: "Generated Title\n" }),
-      LLMEvent.stepFinish({
-        index: 0,
-        reason: { normalized: "stop" },
-        usage: {
-          inputTokens: 15,
-          outputTokens: 6,
-          nonCachedInputTokens: 10,
-          cacheReadInputTokens: 3,
-          cacheWriteInputTokens: 2,
-          reasoningTokens: 2,
-        },
-      }),
-      LLMEvent.finish({
-        reason: { normalized: "stop" },
-      }),
-    )
+    return titleStream()
   },
   generate: () => Effect.die("unused"),
 })
@@ -89,7 +92,7 @@ const it = testEffect(
   ),
 )
 
-const insertSession = (id: Session.ID) =>
+const insertSession = (id: Session.ID, title?: string, created?: number) =>
   Effect.gen(function* () {
     const { db } = yield* Database.Service
     yield* db
@@ -105,7 +108,8 @@ const insertSession = (id: Session.ID) =>
         project_id: Project.ID.global,
         slug: id,
         directory: "/project",
-        title: "New session - fake",
+        title,
+        time_created: created,
         version: "test",
       })
       .onConflictDoNothing()
@@ -131,6 +135,7 @@ const prompt = (sessionID: Session.ID, text: string) =>
 it.effect("generates a title from the sole user message and renames the session", () =>
   Effect.gen(function* () {
     requests = []
+    titleStream = successfulTitle
     const agentService = yield* Agent.Service
     yield* agentService.transform((editor) => {
       editor.update(Agent.ID.make("title"), (agent) => {
@@ -144,13 +149,11 @@ it.effect("generates a title from the sole user message and renames the session"
     yield* prompt(sessionID, "Help me debug the failing build")
 
     const store = yield* SessionStore.Service
-    const session = yield* store
-      .get(sessionID)
-      .pipe(Effect.flatMap((session) => (session ? Effect.succeed(session) : Effect.die("session missing"))))
     const title = yield* SessionTitle.Service
-    yield* title.generateForFirstPrompt(session)
+    yield* title.generateForFirstPrompt(sessionID)
 
     expect(requests).toHaveLength(1)
+    expect(requests[0]?.cache).toEqual({ mode: "auto", key: sessionID })
     expect(requests[0]?.http?.headers).toEqual({
       "x-session-affinity": sessionID,
       "X-Session-Id": sessionID,
@@ -167,9 +170,10 @@ it.effect("generates a title from the sole user message and renames the session"
   }),
 )
 
-it.effect("does not generate once a second user message exists", () =>
+it.effect("generates from the first user message after later messages exist", () =>
   Effect.gen(function* () {
     requests = []
+    titleStream = successfulTitle
     const agentService = yield* Agent.Service
     yield* agentService.transform((editor) => {
       editor.update(Agent.ID.make("title"), (agent) => {
@@ -184,21 +188,46 @@ it.effect("does not generate once a second user message exists", () =>
     yield* prompt(sessionID, "Second message")
 
     const store = yield* SessionStore.Service
-    const session = yield* store
-      .get(sessionID)
-      .pipe(Effect.flatMap((session) => (session ? Effect.succeed(session) : Effect.die("session missing"))))
     const title = yield* SessionTitle.Service
-    yield* title.generateForFirstPrompt(session)
+    yield* title.generateForFirstPrompt(sessionID)
 
-    expect(requests).toHaveLength(0)
-    const untouched = yield* store.get(sessionID)
-    expect(untouched?.title).toBe("New session - fake")
+    expect(requests).toHaveLength(1)
+    expect(JSON.stringify(requests[0]?.messages)).toContain("First message")
+    expect(JSON.stringify(requests[0]?.messages)).not.toContain("Second message")
+    expect((yield* store.get(sessionID))?.title).toBe("Generated Title")
+  }),
+)
+
+it.effect("retries a legacy persisted fallback title", () =>
+  Effect.gen(function* () {
+    requests = []
+    titleStream = successfulTitle
+    const agentService = yield* Agent.Service
+    yield* agentService.transform((editor) => {
+      editor.update(Agent.ID.make("title"), (agent) => {
+        agent.mode = "primary"
+        agent.hidden = true
+        agent.system = "You are a title generator."
+      })
+    })
+    const sessionID = Session.ID.make("ses_title_legacy")
+    const created = Date.parse("2026-07-30T18:45:03.662Z")
+    yield* insertSession(sessionID, "New session - 2026-07-30T18:45:03.662Z", created)
+    yield* prompt(sessionID, "Retry the legacy title")
+
+    const title = yield* SessionTitle.Service
+    yield* title.generateForFirstPrompt(sessionID)
+
+    const store = yield* SessionStore.Service
+    expect(requests).toHaveLength(1)
+    expect((yield* store.get(sessionID))?.title).toBe("Generated Title")
   }),
 )
 
 it.effect("does not generate for a child session", () =>
   Effect.gen(function* () {
     requests = []
+    titleStream = successfulTitle
     const agentService = yield* Agent.Service
     yield* agentService.transform((editor) => {
       editor.update(Agent.ID.make("title"), (agent) => {
@@ -223,7 +252,6 @@ it.effect("does not generate for a child session", () =>
         parent_id: Session.ID.make("ses_title_parent"),
         slug: sessionID,
         directory: "/project",
-        title: "Child session - fake",
         version: "test",
       })
       .onConflictDoNothing()
@@ -231,12 +259,8 @@ it.effect("does not generate for a child session", () =>
       .pipe(Effect.orDie)
     yield* prompt(sessionID, "Do this subtask")
 
-    const store = yield* SessionStore.Service
-    const session = yield* store
-      .get(sessionID)
-      .pipe(Effect.flatMap((session) => (session ? Effect.succeed(session) : Effect.die("session missing"))))
     const title = yield* SessionTitle.Service
-    yield* title.generateForFirstPrompt(session)
+    yield* title.generateForFirstPrompt(sessionID)
 
     expect(requests).toHaveLength(0)
   }),
@@ -245,19 +269,100 @@ it.effect("does not generate for a child session", () =>
 it.effect("does not generate when the title agent is removed", () =>
   Effect.gen(function* () {
     requests = []
+    titleStream = successfulTitle
     const sessionID = Session.ID.make("ses_title_no_agent")
     yield* insertSession(sessionID)
     yield* prompt(sessionID, "Help me debug the failing build")
 
     const store = yield* SessionStore.Service
-    const session = yield* store
-      .get(sessionID)
-      .pipe(Effect.flatMap((session) => (session ? Effect.succeed(session) : Effect.die("session missing"))))
     const title = yield* SessionTitle.Service
-    yield* title.generateForFirstPrompt(session)
+    yield* title.generateForFirstPrompt(sessionID)
 
     expect(requests).toHaveLength(0)
     const untouched = yield* store.get(sessionID)
-    expect(untouched?.title).toBe("New session - fake")
+    expect(untouched?.title).toBeUndefined()
+  }),
+)
+
+it.effect("does not overwrite an explicit title", () =>
+  Effect.gen(function* () {
+    requests = []
+    titleStream = successfulTitle
+    const sessionID = Session.ID.make("ses_title_explicit")
+    yield* insertSession(sessionID)
+    yield* prompt(sessionID, "Help me debug the failing build")
+    const events = yield* Bus.Service
+    yield* events.publish(SessionEvent.Renamed, { sessionID, title: "New session - 2099-01-01T00:00:00.000Z" })
+
+    const title = yield* SessionTitle.Service
+    yield* title.generateForFirstPrompt(sessionID)
+
+    const store = yield* SessionStore.Service
+    expect(requests).toHaveLength(0)
+    expect((yield* store.get(sessionID))?.title).toBe("New session - 2099-01-01T00:00:00.000Z")
+  }),
+)
+
+it.effect("retries after a failed title request", () =>
+  Effect.gen(function* () {
+    requests = []
+    const agentService = yield* Agent.Service
+    yield* agentService.transform((editor) => {
+      editor.update(Agent.ID.make("title"), (agent) => {
+        agent.mode = "primary"
+        agent.hidden = true
+        agent.system = "You are a title generator."
+      })
+    })
+    const sessionID = Session.ID.make("ses_title_retry")
+    yield* insertSession(sessionID)
+    yield* prompt(sessionID, "Retry this title")
+    const title = yield* SessionTitle.Service
+    titleStream = () => Stream.make(LLMEvent.providerError({ message: "Provider unavailable" }))
+
+    yield* title.generateForFirstPrompt(sessionID)
+    titleStream = successfulTitle
+    yield* title.generateForFirstPrompt(sessionID)
+
+    const store = yield* SessionStore.Service
+    expect(requests).toHaveLength(2)
+    expect((yield* store.get(sessionID))?.title).toBe("Generated Title")
+  }),
+)
+
+it.effect("preserves a manual rename completed while generation is in flight", () =>
+  Effect.gen(function* () {
+    requests = []
+    const agentService = yield* Agent.Service
+    yield* agentService.transform((editor) => {
+      editor.update(Agent.ID.make("title"), (agent) => {
+        agent.mode = "primary"
+        agent.hidden = true
+        agent.system = "You are a title generator."
+      })
+    })
+    const sessionID = Session.ID.make("ses_title_manual_rename")
+    yield* insertSession(sessionID)
+    yield* prompt(sessionID, "Generate this title")
+    const started = yield* Deferred.make<void>()
+    const release = yield* Deferred.make<void>()
+    titleStream = () =>
+      Stream.unwrap(
+        Deferred.succeed(started, undefined).pipe(
+          Effect.andThen(Deferred.await(release)),
+          Effect.as(successfulTitle()),
+        ),
+      )
+    const title = yield* SessionTitle.Service
+    const fiber = yield* title.generateForFirstPrompt(sessionID).pipe(Effect.forkScoped)
+    yield* Deferred.await(started)
+    const events = yield* Bus.Service
+    yield* events.publish(SessionEvent.Renamed, { sessionID, title: "Manual title" })
+    yield* Deferred.succeed(release, undefined)
+    yield* Fiber.join(fiber)
+
+    const store = yield* SessionStore.Service
+    expect(requests).toHaveLength(1)
+    expect((yield* store.get(sessionID))?.title).toBe("Manual title")
   }),
 )

packages/core/test/tool-read.test.ts

@@ -2,7 +2,7 @@ import { beforeEach, describe, expect } from "bun:test"
 import path from "path"
 import { Effect, Exit, Layer, PlatformError, Stream } from "effect"
 import { Config } from "@opencode-ai/core/config"
-import { ConfigAttachments } from "@opencode-ai/core/config/attachments"
+import { ConfigMedia } from "@opencode-ai/core/config/media"
 import { AppNodeBuilder } from "@opencode-ai/core/effect/app-node-builder"
 import { LayerNode } from "@opencode-ai/util/effect/layer-node"
 import { FileSystem } from "@opencode-ai/core/filesystem"
@@ -432,8 +432,8 @@ describe("ReadTool", () => {
         new Config.Document({
           type: "document",
           info: new Config.Info({
-            attachments: new ConfigAttachments.Info({
-              image: new ConfigAttachments.Image({ auto_resize: false, max_width: 4 }),
+            media: new ConfigMedia.Info({
+              image: new ConfigMedia.Image({ auto_resize: false, max_width: 4 }),
             }),
           }),
         }),
@@ -475,7 +475,7 @@ describe("ReadTool", () => {
         new Config.Document({
           type: "document",
           info: new Config.Info({
-            attachments: new ConfigAttachments.Info({ image: new ConfigAttachments.Image({ max_width: 4 }) }),
+            media: new ConfigMedia.Info({ image: new ConfigMedia.Image({ max_width: 4 }) }),
           }),
         }),
       ])
@@ -514,8 +514,8 @@ describe("ReadTool", () => {
         new Config.Document({
           type: "document",
           info: new Config.Info({
-            attachments: new ConfigAttachments.Info({
-              image: new ConfigAttachments.Image({ max_base64_bytes: 1 }),
+            media: new ConfigMedia.Info({
+              image: new ConfigMedia.Image({ max_base64_bytes: 1 }),
             }),
           }),
         }),

packages/core/test/tool-subagent.test.ts

@@ -47,7 +47,7 @@ const executionNode = makeGlobalNode({
       const completed = new Set<Session.ID>()
       const complete = Effect.fn("SubagentTest.complete")(function* (sessionID: Session.ID) {
         if (completed.has(sessionID)) return
-        if ((yield* store.get(sessionID))?.title.includes("fail")) {
+        if ((yield* store.get(sessionID))?.title?.includes("fail")) {
           yield* new SessionRunnerModel.ModelNotSelectedError({ sessionID })
           return
         }

packages/enterprise/package.json

@@ -16,6 +16,7 @@
     "@opencode-ai/core": "workspace:*",
     "@opencode-ai/session-ui": "workspace:*",
     "@opencode-ai/ui": "workspace:*",
+    "@opencode-ai/util": "workspace:*",
     "aws4fetch": "^1.0.20",
     "@pierre/diffs": "catalog:",
     "@solidjs/router": "catalog:",

packages/enterprise/src/routes/share/[shareID].tsx

@@ -4,6 +4,7 @@ import { SessionReview } from "@opencode-ai/session-ui/session-review"
 import { DataProvider } from "@opencode-ai/session-ui/context"
 import { FileComponentProvider } from "@opencode-ai/ui/context/file"
 import { WorkerPoolProvider } from "@opencode-ai/ui/context/worker-pool"
+import { withTimestampedFallback } from "@opencode-ai/util/session-title-fallback"
 import { createAsync, query, useParams } from "@solidjs/router"
 import { createMemo, createSignal, ErrorBoundary, For, Match, Show, Switch } from "solid-js"
 import { Share } from "~/core/share"
@@ -158,6 +159,7 @@ export default function () {
           const match = createMemo(() => Binary.search(data().session, data().sessionID, (s) => s.id))
           if (!match().found) throw new Error(`Session ${data().sessionID} not found`)
           const info = createMemo(() => data().session[match().index])
+          const title = createMemo(() => withTimestampedFallback(info()))
           const ogImage = createMemo(() => {
             const models = new Set<string>()
             const messages = data().message[data().sessionID] ?? []
@@ -167,7 +169,7 @@ export default function () {
               }
             }
             const modelIDs = Array.from(models)
-            const encodedTitle = encodeURIComponent(Base64.encode(encodeURIComponent(info().title.substring(0, 700))))
+            const encodedTitle = encodeURIComponent(Base64.encode(encodeURIComponent(title().substring(0, 700))))
             let modelParam: string
             if (modelIDs.length === 1) {
               modelParam = modelIDs[0]
@@ -184,9 +186,7 @@ export default function () {
 
           return (
             <>
-              <Show when={info().title}>
-                <Title>{info().title} | OpenCode</Title>
-              </Show>
+              <Title>{title()} | OpenCode</Title>
               <Meta name="description" content="opencode - The AI coding agent built for the terminal." />
               <Meta property="og:image" content={ogImage()} />
               <Meta name="twitter:image" content={ogImage()} />
@@ -240,7 +240,7 @@ export default function () {
                               </div>
                             </div>
                           </div>
-                          <div class="text-left text-16-medium text-text-strong">{info().title}</div>
+                          <div class="text-left text-16-medium text-text-strong">{title()}</div>
                         </div>
                       )
 

packages/plugin/src/effect/session.ts

@@ -1,5 +1,6 @@
 import type { SessionApi } from "@opencode-ai/client/effect/api"
 import type { Message, SystemPart } from "@opencode-ai/ai"
+import type { HttpRequest } from "@opencode-ai/ai/route"
 import type { Agent } from "@opencode-ai/schema/agent"
 import type { Model } from "@opencode-ai/schema/model"
 import type { Session } from "@opencode-ai/schema/session"
@@ -15,13 +16,20 @@ export interface SessionContext {
   tools: Record<string, { description: string; input: JsonSchema.JsonSchema }>
 }
 
+export interface SessionRequest extends HttpRequest {
+  readonly sessionID: Session.ID
+  readonly agent: Agent.ID
+  readonly model: Model.Ref
+}
+
 export interface SessionHooks {
   readonly context: SessionContext
+  readonly request: SessionRequest
 }
 
 export type SessionDomain = Pick<
   SessionApi<unknown>,
-  "create" | "get" | "prompt" | "generate" | "command" | "synthetic" | "interrupt"
+  "create" | "get" | "prompt" | "generate" | "command" | "synthetic" | "interrupt" | "rename" | "wait"
 > & {
   readonly hook: Hooks<SessionHooks>
 }

packages/plugin/src/promise/session.ts

@@ -1,5 +1,6 @@
 import type { SessionApi } from "@opencode-ai/client/promise/api"
 import type { Message, SystemPart } from "@opencode-ai/ai"
+import type { HttpRequest } from "@opencode-ai/ai/route"
 import type { Agent } from "@opencode-ai/schema/agent"
 import type { Model } from "@opencode-ai/schema/model"
 import type { Session } from "@opencode-ai/schema/session"
@@ -15,8 +16,15 @@ export interface SessionContext {
   tools: Record<string, { description: string; input: JsonSchema.JsonSchema }>
 }
 
+export interface SessionRequest extends HttpRequest {
+  readonly sessionID: Session.ID
+  readonly agent: Agent.ID
+  readonly model: Model.Ref
+}
+
 export interface SessionHooks {
   readonly context: SessionContext
+  readonly request: SessionRequest
 }
 
 export type SessionDomain = Pick<

packages/plugin/src/tui/context.ts

@@ -26,12 +26,27 @@ import type { JSX } from "@opentui/solid"
 import type { Store } from "solid-js/store"
 
 export interface Storage {
+  /**
+   * Durable JSON state: persisted to disk, survives hot reloads and TUI
+   * restarts, and stays live-synced across running TUI instances.
+   */
   store<Value extends object>(
     key: string,
     options: {
       readonly initial: Value
     },
   ): readonly [Store<Value>, (mutation: (draft: Value) => void) => Promise<void>]
+  /**
+   * Ephemeral in-memory state: survives plugin hot reloads (old and new
+   * generations share the same live store) and is gone when the TUI exits.
+   * Updates are synchronous and values need not be JSON-serializable.
+   */
+  memory<Value extends object>(
+    key: string,
+    options: {
+      readonly initial: Value
+    },
+  ): readonly [Store<Value>, (mutation: (draft: Value) => void) => void]
 }
 
 interface LocationCollection<Value> {
@@ -136,6 +151,9 @@ export interface SlotMap {
     readonly sessionID?: string
     readonly mode: "normal" | "shell"
   }
+  readonly "session.composer.top": {
+    readonly sessionID: string
+  }
   readonly "sidebar.content": {
     readonly sessionID: string
   }

packages/protocol/src/groups/session.ts

@@ -149,6 +149,7 @@ export const makeSessionGroup = <I extends HttpApiMiddleware.AnyId, S>(sessionLo
       HttpApiEndpoint.post("session.create", "/api/session", {
         payload: Schema.Struct({
           id: Session.ID.pipe(Schema.optional),
+          title: Schema.String.pipe(Schema.optional),
           agent: Agent.ID.pipe(Schema.optional),
           model: Model.Ref.pipe(Schema.optional),
           location: Location.Ref.pipe(Schema.optional),

packages/schema/src/session.ts

@@ -42,7 +42,7 @@ export const Info = Schema.Struct({
     updated: DateTimeUtcFromMillis,
     archived: DateTimeUtcFromMillis.pipe(optional),
   }),
-  title: Schema.String,
+  title: Schema.String.pipe(optional),
   location: Location.Ref,
   subpath: RelativePath.pipe(optional),
   revert: Revert.pipe(optional),

packages/schema/src/v1/session.ts

@@ -552,7 +552,7 @@ export const SessionInfo = Schema.Struct({
   cost: optional(Schema.Finite),
   tokens: optional(SessionTokens),
   share: optional(SessionShare),
-  title: Schema.String,
+  title: optional(Schema.String),
   agent: optional(Schema.String),
   model: optional(SessionModel),
   version: Schema.String,