feat(providers): Google Gemini as its own backend with thought signatures

Google joins as a first-class provider (kind `google`, aliases
`gemini`/`google-gemini`/`ai-studio`) on the official Gemini
OpenAI-compatible endpoint
(generativelanguage.googleapis.com/v1beta/openai/), keyed by
GOOGLE_API_KEY / GEMINI_API_KEY. It is a distinct backend, not an
OpenAI alias.

Thought signatures, required for tool-call replay on that route:

- Capture: non-streaming responses and the first streaming chunk of
  each tool call carry extra_content.google.thought_signature; both
  paths store it on the ToolUse content block and it survives the
  stream-to-message assembly.
- Replay: assistant tool-call messages re-attach the signature as
  extra_content.google.thought_signature on the exact official route
  only. Every other provider - and a google row pointed at another
  gateway - gets the field stripped so nothing Google-specific leaks.
- Fail closed: replaying a tool call whose signature was not captured
  on a thinking model (gemini-3*, gemini-2.5-pro/flash) fails before
  transport with an actionable error. gemini-2.5-flash-lite ships
  thinking off and degrades with a warning instead.

Reasoning effort maps onto the documented google.thinking_config.
thinking_level body field (low/high) on the exact route.

Eight focused tests prove round-trip, fail-closed, flash-lite
degradation, cross-provider stripping, neighbor-gateway neutrality,
thinking-level mapping, and both capture paths. Provider registries,
picker tables, CLI arg, agent model roster, and PROVIDERS.md rows are
wired (39 catalog kinds / 44 registry entries).

WIP against the live endpoint: no Gemini key exists in this
environment, so live calls are UNRUN; behavior is verified offline
only.

crates/agent/src/lib.rs

@@ -1196,6 +1196,55 @@ impl Default for ModelRegistry {
                 supports_tools: true,
                 supports_reasoning: false,
             },
+            ModelInfo {
+                id: "gemini-3.1-pro-preview".to_string(),
+                provider: ProviderKind::Google,
+                aliases: vec!["gemini-3.1-pro".to_string()],
+                supports_tools: true,
+                supports_reasoning: true,
+            },
+            ModelInfo {
+                id: "gemini-3-pro-preview".to_string(),
+                provider: ProviderKind::Google,
+                aliases: vec!["gemini-3-pro".to_string()],
+                supports_tools: true,
+                supports_reasoning: true,
+            },
+            ModelInfo {
+                id: "gemini-3.6-flash".to_string(),
+                provider: ProviderKind::Google,
+                aliases: vec![],
+                supports_tools: true,
+                supports_reasoning: true,
+            },
+            ModelInfo {
+                id: "gemini-3.5-flash".to_string(),
+                provider: ProviderKind::Google,
+                aliases: vec![],
+                supports_tools: true,
+                supports_reasoning: true,
+            },
+            ModelInfo {
+                id: "gemini-3.5-flash-lite".to_string(),
+                provider: ProviderKind::Google,
+                aliases: vec![],
+                supports_tools: true,
+                supports_reasoning: false,
+            },
+            ModelInfo {
+                id: "gemini-2.5-pro".to_string(),
+                provider: ProviderKind::Google,
+                aliases: vec![],
+                supports_tools: true,
+                supports_reasoning: true,
+            },
+            ModelInfo {
+                id: "gemini-2.5-flash".to_string(),
+                provider: ProviderKind::Google,
+                aliases: vec![],
+                supports_tools: true,
+                supports_reasoning: true,
+            },
             ModelInfo {
                 id: "mistral-code-latest".to_string(),
                 provider: ProviderKind::Mistral,

crates/cli/src/lib.rs

@@ -108,6 +108,8 @@ enum ProviderArg {
         alias = "la_plateforme"
     )]
     Mistral,
+    /// Google Gemini (official OpenAI-compatible endpoint).
+    Google,
 }
 
 impl From<ProviderArg> for ProviderKind {
@@ -148,6 +150,7 @@ impl From<ProviderArg> for ProviderKind {
             ProviderArg::Meta => ProviderKind::Meta,
             ProviderArg::Xai => ProviderKind::Xai,
             ProviderArg::Mistral => ProviderKind::Mistral,
+            ProviderArg::Google => ProviderKind::Google,
         }
     }
 }

crates/config/src/lib.rs

@@ -412,6 +412,16 @@ pub struct ProvidersToml {
         alias = "la_plateforme"
     )]
     pub mistral: ProviderConfigToml,
+    /// Google Gemini — official OpenAI-compatible endpoint with thought
+    /// signatures on tool calls.
+    #[serde(
+        default,
+        skip_serializing_if = "ProviderConfigToml::is_empty",
+        alias = "google-gemini",
+        alias = "google_gemini",
+        alias = "gemini"
+    )]
+    pub google: ProviderConfigToml,
     /// Jiangsu Telecom TokenHub — OpenAI-compatible AI gateway.
     #[serde(
         default,
@@ -641,6 +651,7 @@ impl ProvidersToml {
             ProviderKind::Meta => &self.meta,
             ProviderKind::Xai => &self.xai,
             ProviderKind::Mistral => &self.mistral,
+            ProviderKind::Google => &self.google,
             ProviderKind::Telecomjs => &self.telecomjs,
             ProviderKind::ModelstudioTokenPlan => &self.modelstudio_token_plan,
             ProviderKind::ModelstudioTokenPlanAnthropic => &self.modelstudio_token_plan_anthropic,
@@ -689,6 +700,7 @@ impl ProvidersToml {
             ProviderKind::Meta => &mut self.meta,
             ProviderKind::Xai => &mut self.xai,
             ProviderKind::Mistral => &mut self.mistral,
+            ProviderKind::Google => &mut self.google,
             ProviderKind::Telecomjs => &mut self.telecomjs,
             ProviderKind::ModelstudioTokenPlan => &mut self.modelstudio_token_plan,
             ProviderKind::ModelstudioTokenPlanAnthropic => {
@@ -3209,6 +3221,7 @@ impl ConfigToml {
                 ProviderKind::Meta => DEFAULT_META_BASE_URL.to_string(),
                 ProviderKind::Xai => DEFAULT_XAI_BASE_URL.to_string(),
                 ProviderKind::Mistral => DEFAULT_MISTRAL_BASE_URL.to_string(),
+                ProviderKind::Google => DEFAULT_GOOGLE_BASE_URL.to_string(),
                 ProviderKind::Telecomjs => DEFAULT_TELECOMJS_BASE_URL.to_string(),
                 ProviderKind::ModelstudioTokenPlan
                 | ProviderKind::ModelstudioTokenPlanAnthropic
@@ -4300,6 +4313,7 @@ fn default_model_for_provider(provider: ProviderKind) -> &'static str {
         ProviderKind::Meta => DEFAULT_META_MODEL,
         ProviderKind::Xai => DEFAULT_XAI_MODEL,
         ProviderKind::Mistral => DEFAULT_MISTRAL_MODEL,
+        ProviderKind::Google => DEFAULT_GOOGLE_MODEL,
         ProviderKind::Telecomjs => DEFAULT_TELECOMJS_MODEL,
         ProviderKind::ModelstudioTokenPlan
         | ProviderKind::ModelstudioTokenPlanAnthropic
@@ -4349,6 +4363,7 @@ fn default_base_url_for_provider(provider: ProviderKind) -> &'static str {
         ProviderKind::Meta => DEFAULT_META_BASE_URL,
         ProviderKind::Xai => DEFAULT_XAI_BASE_URL,
         ProviderKind::Mistral => DEFAULT_MISTRAL_BASE_URL,
+        ProviderKind::Google => DEFAULT_GOOGLE_BASE_URL,
         ProviderKind::Telecomjs => DEFAULT_TELECOMJS_BASE_URL,
         ProviderKind::ModelstudioTokenPlan => DEFAULT_MODELSTUDIO_TOKEN_PLAN_BASE_URL,
         ProviderKind::ModelstudioTokenPlanAnthropic => MODELSTUDIO_TOKEN_PLAN_ANTHROPIC_BASE_URL,
@@ -6608,6 +6623,8 @@ struct EnvRuntimeOverrides {
     xai_model: Option<String>,
     mistral_base_url: Option<String>,
     mistral_model: Option<String>,
+    google_base_url: Option<String>,
+    google_model: Option<String>,
     telecomjs_base_url: Option<String>,
     telecomjs_model: Option<String>,
     modelstudio_token_plan_base_url: Option<String>,
@@ -6918,6 +6935,22 @@ impl EnvRuntimeOverrides {
             xai_model: std::env::var("XAI_MODEL")
                 .ok()
                 .filter(|v| !v.trim().is_empty()),
+            google_base_url: std::env::var("GOOGLE_BASE_URL")
+                .ok()
+                .filter(|v| !v.trim().is_empty())
+                .or_else(|| {
+                    std::env::var("GEMINI_BASE_URL")
+                        .ok()
+                        .filter(|v| !v.trim().is_empty())
+                }),
+            google_model: std::env::var("GOOGLE_MODEL")
+                .ok()
+                .filter(|v| !v.trim().is_empty())
+                .or_else(|| {
+                    std::env::var("GEMINI_MODEL")
+                        .ok()
+                        .filter(|v| !v.trim().is_empty())
+                }),
             mistral_base_url: std::env::var("MISTRAL_BASE_URL")
                 .ok()
                 .filter(|v| !v.trim().is_empty()),
@@ -7026,6 +7059,7 @@ impl EnvRuntimeOverrides {
             ProviderKind::Meta => self.meta_base_url.clone(),
             ProviderKind::Xai => self.xai_base_url.clone(),
             ProviderKind::Mistral => self.mistral_base_url.clone(),
+            ProviderKind::Google => self.google_base_url.clone(),
             ProviderKind::Telecomjs => self.telecomjs_base_url.clone(),
             ProviderKind::ModelstudioTokenPlan | ProviderKind::ModelstudioTokenPlanAnthropic => {
                 self.modelstudio_token_plan_base_url.clone()
@@ -7070,6 +7104,7 @@ impl EnvRuntimeOverrides {
             ProviderKind::Meta => self.meta_model.clone(),
             ProviderKind::Xai => self.xai_model.clone(),
             ProviderKind::Mistral => self.mistral_model.clone(),
+            ProviderKind::Google => self.google_model.clone(),
             ProviderKind::Telecomjs => self.telecomjs_model.clone(),
             ProviderKind::ModelstudioTokenPlan | ProviderKind::ModelstudioTokenPlanAnthropic => {
                 self.modelstudio_token_plan_model.clone()

crates/config/src/provider.rs

@@ -9,8 +9,9 @@ use super::{
     DEFAULT_ATLASCLOUD_MODEL, DEFAULT_DEEPINFRA_BASE_URL, DEFAULT_DEEPINFRA_MODEL,
     DEFAULT_DEEPSEEK_ANTHROPIC_BASE_URL, DEFAULT_DEEPSEEK_ANTHROPIC_MODEL,
     DEFAULT_DEEPSEEK_BASE_URL, DEFAULT_DEEPSEEK_MODEL, DEFAULT_FIREWORKS_BASE_URL,
-    DEFAULT_FIREWORKS_MODEL, DEFAULT_HUGGINGFACE_BASE_URL, DEFAULT_HUGGINGFACE_MODEL,
-    DEFAULT_LONGCAT_BASE_URL, DEFAULT_LONGCAT_MODEL, DEFAULT_META_BASE_URL, DEFAULT_META_MODEL,
+    DEFAULT_FIREWORKS_MODEL, DEFAULT_GOOGLE_BASE_URL, DEFAULT_GOOGLE_MODEL,
+    DEFAULT_HUGGINGFACE_BASE_URL, DEFAULT_HUGGINGFACE_MODEL, DEFAULT_LONGCAT_BASE_URL,
+    DEFAULT_LONGCAT_MODEL, DEFAULT_META_BASE_URL, DEFAULT_META_MODEL,
     DEFAULT_MINIMAX_ANTHROPIC_BASE_URL, DEFAULT_MINIMAX_BASE_URL, DEFAULT_MINIMAX_MODEL,
     DEFAULT_MISTRAL_BASE_URL, DEFAULT_MISTRAL_MODEL, DEFAULT_MODELSTUDIO_CODING_PLAN_BASE_URL,
     DEFAULT_MODELSTUDIO_TOKEN_PLAN_BASE_URL, DEFAULT_MODELSTUDIO_TOKEN_PLAN_MODEL,
@@ -420,6 +421,12 @@ pub const fn credential_help(kind: ProviderKind) -> CredentialHelp {
             docs_url: Some("https://www.alibabacloud.com/help/en/model-studio/"),
             guidance: "Sign in to Alibaba Cloud Model Studio (Bailian console), create or copy an API key, and select the plan endpoint matching your subscription (Token Plan or Coding Plan).",
         },
+        ProviderKind::Google => CredentialHelp {
+            acquisition: ApiKey,
+            credential_url: Some("https://aistudio.google.com/apikey"),
+            docs_url: Some("https://ai.google.dev/gemini-api/docs/openai"),
+            guidance: "Create a Google AI Studio API key. Codewhale uses the official Gemini OpenAI-compatible endpoint and never reads Google OAuth files.",
+        },
         ProviderKind::Custom => CredentialHelp {
             acquisition: Configuration,
             credential_url: None,
@@ -952,6 +959,18 @@ provider!(
     aliases: ["mistral-ai", "mistral_ai", "mistralai", "la-plateforme", "la_plateforme"]
 );
 
+provider!(
+    Google,
+    Google,
+    "google",
+    "Google Gemini",
+    DEFAULT_GOOGLE_BASE_URL,
+    DEFAULT_GOOGLE_MODEL,
+    ["GOOGLE_API_KEY", "GEMINI_API_KEY"],
+    "google",
+    aliases: ["google-gemini", "google_gemini", "gemini", "google-ai", "google_ai", "ai-studio", "aistudio"]
+);
+
 /// OpenAI Codex / ChatGPT OAuth provider using the Responses API.
 pub struct OpenaiCodex;
 
@@ -1580,7 +1599,7 @@ static MODELSTUDIO_CODING_PLAN_ANTHROPIC: ModelstudioCodingPlanAnthropic =
     ModelstudioCodingPlanAnthropic;
 static CUSTOM: Custom = Custom;
 
-static PROVIDER_REGISTRY: [&dyn Provider; 43] = [
+static PROVIDER_REGISTRY: [&dyn Provider; 44] = [
     &DEEPSEEK,
     &DEEPSEEK_ANTHROPIC,
     &NVIDIA_NIM,
@@ -1623,6 +1642,7 @@ static PROVIDER_REGISTRY: [&dyn Provider; 43] = [
     &MODELSTUDIO_TOKEN_PLAN_ANTHROPIC,
     &MODELSTUDIO_CODING_PLAN,
     &MODELSTUDIO_CODING_PLAN_ANTHROPIC,
+    &Google,
     &CUSTOM,
 ];
 

crates/config/src/provider_defaults.rs

@@ -190,3 +190,9 @@ pub(crate) const DEFAULT_MODELSTUDIO_CODING_PLAN_BASE_URL: &str =
     "https://coding-intl.dashscope.aliyuncs.com/v1";
 pub(crate) const MODELSTUDIO_CODING_PLAN_ANTHROPIC_BASE_URL: &str =
     "https://coding-intl.dashscope.aliyuncs.com/apps/anthropic";
+
+/// Google Gemini OpenAI-compatible Chat Completions base URL.
+pub const DEFAULT_GOOGLE_BASE_URL: &str =
+    "https://generativelanguage.googleapis.com/v1beta/openai/";
+/// Default Gemini model for the Google provider (preview flagship, 2026-08).
+pub const DEFAULT_GOOGLE_MODEL: &str = "gemini-3.1-pro-preview";

crates/config/src/provider_kind.rs

@@ -189,6 +189,19 @@ pub enum ProviderKind {
         alias = "alibaba-coding-plan-anthropic"
     )]
     ModelstudioCodingPlanAnthropic,
+    /// Google — Gemini OpenAI-compatible endpoint. Its own backend, not an
+    /// OpenAI alias: thought signatures on tool calls are captured and
+    /// replayed per Google's contract.
+    #[serde(
+        alias = "google-gemini",
+        alias = "google_gemini",
+        alias = "gemini",
+        alias = "google-ai",
+        alias = "google_ai",
+        alias = "ai-studio",
+        alias = "aistudio"
+    )]
+    Google,
     /// User-defined OpenAI-compatible endpoint (#1519).
     ///
     /// A single dynamic identity for arbitrary `[providers.<name>]
@@ -206,7 +219,7 @@ impl ProviderKind {
     /// stay on the enum for serde and `provider_for_kind`, but they are not
     /// first-class catalog rows. Plan is `mode` / base_url; dialect is
     /// `wire = openai|anthropic` on the primary provider config.
-    pub const ALL: [Self; 38] = [
+    pub const ALL: [Self; 39] = [
         Self::Deepseek,
         Self::NvidiaNim,
         Self::Openai,
@@ -244,6 +257,7 @@ impl ProviderKind {
         Self::Mistral,
         Self::Telecomjs,
         Self::ModelstudioTokenPlan,
+        Self::Google,
         Self::Custom,
     ];
 

crates/config/src/tests.rs

@@ -4811,9 +4811,9 @@ fn meta_model_api_scopes_both_documented_key_names_to_official_endpoint() {
 fn provider_metadata_registry_covers_every_provider_kind_once() {
     let providers = provider::all_providers();
     // Full registry keeps legacy dialect/plan kinds for provider_for_kind.
-    assert_eq!(providers.len(), 43);
+    assert_eq!(providers.len(), 44);
     // Catalog surface is one identity per vendor (no dual-wire / plan rows).
-    assert_eq!(ProviderKind::ALL.len(), 38);
+    assert_eq!(ProviderKind::ALL.len(), 39);
     assert!(ProviderKind::ALL.len() < providers.len());
 
     let mut ids = std::collections::BTreeSet::new();

crates/core/src/request.rs

@@ -158,6 +158,12 @@ pub enum ContentBlock {
         input: serde_json::Value,
         #[serde(skip_serializing_if = "Option::is_none")]
         caller: Option<ToolCaller>,
+        /// Google thought signature captured from the OpenAI-compat route's
+        /// `extra_content.google.thought_signature` on the tool call. Google
+        /// requires replaying it with the tool result for thinking models;
+        /// skipped on the wire and in storage for every other provider.
+        #[serde(skip_serializing_if = "Option::is_none", default)]
+        thought_signature: Option<String>,
     },
     #[serde(rename = "tool_result")]
     ToolResult {

crates/tui/src/acp_server.rs

@@ -457,7 +457,7 @@ where
                         match event {
                             StreamEvent::ContentBlockStart {
                                 index,
-                                content_block: ContentBlockStart::ToolUse { id, name, input, .. },
+                                content_block: ContentBlockStart::ToolUse { id, name, input, ..},
                             } => {
                                 pending_tool_uses.insert(
                                     index,
@@ -1313,6 +1313,7 @@ where
                 name: call.name.clone(),
                 input: call.input.clone(),
                 caller: None,
+                thought_signature: None,
             });
         }
         if !assistant_content.is_empty() {
@@ -2652,6 +2653,7 @@ mod tests {
                     name: name.to_string(),
                     input: json!({}),
                     caller: None,
+                    thought_signature: None,
                 },
             },
             StreamEvent::ContentBlockDelta {

crates/tui/src/client.rs

@@ -3239,6 +3239,7 @@ pub(super) fn apply_reasoning_effort(
             ApiProvider::Meta => {}
             ApiProvider::Xai => {}
             ApiProvider::Mistral => {}
+            ApiProvider::Google => {}
         },
         "low" | "minimal" | "medium" | "mid" | "high" | "" => match provider {
             // Handled by the shared DeepSeek table above, before this match.
@@ -3348,6 +3349,7 @@ pub(super) fn apply_reasoning_effort(
             ApiProvider::Meta => {}
             ApiProvider::Xai => {}
             ApiProvider::Mistral => {}
+            ApiProvider::Google => {}
         },
         "xhigh" | "max" | "highest" | "ultracode" => match provider {
             // Handled by the shared DeepSeek table above, before this match.
@@ -3434,6 +3436,7 @@ pub(super) fn apply_reasoning_effort(
             ApiProvider::Meta => {}
             ApiProvider::Xai => {}
             ApiProvider::Mistral => {}
+            ApiProvider::Google => {}
         },
         _ => {}
     }
@@ -4784,6 +4787,7 @@ mod tests {
                         name: "read_file".to_string(),
                         input: json!({"path": "src/lib.rs"}),
                         caller: None,
+                        thought_signature: None,
                     },
                 ],
             },
@@ -5545,6 +5549,7 @@ mod tests {
                         name: "read_file".to_string(),
                         input: json!({"path": "config.toml"}),
                         caller: None,
+                        thought_signature: None,
                     }],
                 },
                 Message {
@@ -6877,6 +6882,7 @@ mod tests {
                         name: "get_date".to_string(),
                         input: json!({}),
                         caller: None,
+                        thought_signature: None,
                     },
                 ],
             },
@@ -6930,6 +6936,7 @@ mod tests {
                         name: "get_date".to_string(),
                         input: json!({}),
                         caller: None,
+                        thought_signature: None,
                     },
                 ],
             },
@@ -7101,6 +7108,7 @@ mod tests {
                         name: "read_file".to_string(),
                         input: json!({"path": "Cargo.toml"}),
                         caller: None,
+                        thought_signature: None,
                     }],
                 },
                 Message {
@@ -8324,6 +8332,7 @@ mod tests {
                         name: "list_dir".to_string(),
                         input: json!({}),
                         caller: None,
+                        thought_signature: None,
                     },
                 ],
             },
@@ -8366,6 +8375,7 @@ mod tests {
                         name: "web.run".to_string(),
                         input: json!({}),
                         caller: None,
+                        thought_signature: None,
                     },
                 ],
             },
@@ -8411,6 +8421,7 @@ mod tests {
                     name: "read_file".to_string(),
                     input: json!({"path": "src/main.rs"}),
                     caller: None,
+                    thought_signature: None,
                 }],
             },
             // No tool result follows — it was removed by compaction.
@@ -8457,6 +8468,7 @@ mod tests {
                         name: "list_dir".to_string(),
                         input: json!({}),
                         caller: None,
+                        thought_signature: None,
                     },
                 ],
             },
@@ -8498,18 +8510,21 @@ mod tests {
                         name: "read_file".to_string(),
                         input: json!({"path": "a.rs"}),
                         caller: None,
+                        thought_signature: None,
                     },
                     ContentBlock::ToolUse {
                         id: "t2".to_string(),
                         name: "read_file".to_string(),
                         input: json!({"path": "b.rs"}),
                         caller: None,
+                        thought_signature: None,
                     },
                     ContentBlock::ToolUse {
                         id: "t3".to_string(),
                         name: "shell".to_string(),
                         input: json!({"cmd": "ls"}),
                         caller: None,
+                        thought_signature: None,
                     },
                 ],
             },

crates/tui/src/client/anthropic.rs

@@ -1176,12 +1176,14 @@ mod tests {
                         name: "read_file".to_string(),
                         input: json!({"path": "a.txt"}),
                         caller: None,
+                        thought_signature: None,
                     },
                     ContentBlock::ToolUse {
                         id: "toolu_orphan".to_string(),
                         name: "task".to_string(),
                         input: json!({}),
                         caller: None,
+                        thought_signature: None,
                     },
                 ],
             },
@@ -1203,6 +1205,7 @@ mod tests {
                     name: "task".to_string(),
                     input: json!({}),
                     caller: None,
+                    thought_signature: None,
                 }],
             },
         ];
@@ -1236,7 +1239,7 @@ mod tests {
         // A fully answered history is left alone.
         request.messages.truncate(3);
         request.messages[1].content.retain(
-            |block| !matches!(block, ContentBlock::ToolUse { id, .. } if id == "toolu_orphan"),
+            |block| !matches!(block, ContentBlock::ToolUse { id, ..} if id == "toolu_orphan"),
         );
         let body = client.build_anthropic_body(&request, true);
         let messages = body["messages"].as_array().expect("messages array");
@@ -1501,6 +1504,7 @@ mod tests {
                         name: "read_file".to_string(),
                         input: json!({"path": "a.txt"}),
                         caller: None,
+                        thought_signature: None,
                     },
                 ],
             },

crates/tui/src/client/chat.rs

@@ -555,6 +555,7 @@ pub(super) fn apply_route_reasoning_controls(
     apply_kimi_code_k3_reasoning_effort(body, provider, base_url, model, effort);
     apply_zai_route_reasoning_controls(body, provider, base_url, model, effort);
     apply_mistral_route_reasoning_controls(body, provider, base_url, model, effort);
+    apply_google_thinking_level(body, provider, base_url, model, effort);
 }
 
 /// Mistral's polymorphic reasoning-content contract is only proven on its
@@ -578,6 +579,124 @@ fn is_exact_mistral_chat_route(provider: ApiProvider, base_url: &str) -> bool {
     ) && path == "v1"
 }
 
+/// Google's OpenAI-compatibility route. Thought signatures are captured
+/// from tool-call `extra_content.google.thought_signature` and replayed on
+/// the assistant tool-call messages of later turns; thinking models fail
+/// closed when a replayed call has no signature.
+fn is_exact_google_chat_route(provider: ApiProvider, base_url: &str) -> bool {
+    if provider != ApiProvider::Google {
+        return false;
+    }
+    let trimmed = base_url.trim().trim_end_matches('/').to_ascii_lowercase();
+    let Some((host, path)) = trimmed
+        .strip_prefix("https://")
+        .and_then(|rest| rest.split_once('/'))
+    else {
+        return false;
+    };
+    host == "generativelanguage.googleapis.com" && path == "v1beta/openai"
+}
+
+/// Gemini models whose thinking makes thought signatures load-bearing on
+/// the OpenAI-compat route. Gemini 2.5 Flash-Lite ships with thinking off
+/// by default, so a missing signature there degrades with a warning
+/// instead of failing the turn.
+fn google_model_requires_thought_signatures(model: &str) -> bool {
+    let model = model.trim().to_ascii_lowercase();
+    if model.starts_with("gemini-3") {
+        return true;
+    }
+    if model.starts_with("gemini-2.5-pro") {
+        return true;
+    }
+    model.starts_with("gemini-2.5-flash") && !model.starts_with("gemini-2.5-flash-lite")
+}
+
+/// Thinking level for the OpenAI-compat route rides the documented
+/// `google.thinking_config.thinking_level` body field (low/high; Gemini 3
+/// cannot disable thinking).
+fn apply_google_thinking_level(
+    body: &mut serde_json::Value,
+    provider: ApiProvider,
+    base_url: &str,
+    _model: &str,
+    effort: Option<&str>,
+) {
+    if !is_exact_google_chat_route(provider, base_url) || effort.is_none() {
+        return;
+    }
+    let level = match effort
+        .unwrap_or_default()
+        .trim()
+        .to_ascii_lowercase()
+        .as_str()
+    {
+        "off" | "disabled" | "none" | "false" | "" | "low" | "minimal" | "medium" | "mid" => "low",
+        _ => "high",
+    };
+    body["google"]["thinking_config"]["thinking_level"] = json!(level);
+}
+
+/// Fail closed before transport when the exact Google route would replay
+/// tool calls without the thought signatures Google's thinking models
+/// require. The error names the model and tells the operator how to
+/// recover instead of letting Google reject or corrupt the tool loop.
+fn validate_google_thought_signature_replay(
+    provider: ApiProvider,
+    base_url: &str,
+    model: &str,
+    messages: &[Value],
+) -> Result<()> {
+    if !is_exact_google_chat_route(provider, base_url)
+        || !google_model_requires_thought_signatures(model)
+    {
+        return Ok(());
+    }
+    for message in messages {
+        let Some(tool_calls) = message.get("tool_calls").and_then(Value::as_array) else {
+            continue;
+        };
+        for call in tool_calls {
+            let missing = call
+                .pointer("/extra_content/google/thought_signature")
+                .and_then(Value::as_str)
+                .is_none();
+            if missing {
+                let id = call.get("id").and_then(Value::as_str).unwrap_or("?");
+                anyhow::bail!(
+                    "Gemini model `{model}` requires a thought signature to replay tool call \
+                     `{id}`, but none was captured (the turn predates signature capture, or \
+                     the provider omitted it). Start a new session before using tools on \
+                     this route."
+                );
+            }
+        }
+    }
+    Ok(())
+}
+
+/// Captured Google signatures ride on tool calls as
+/// `extra_content.google.thought_signature`. Only the exact Google route
+/// may carry them on the wire; every other provider gets them stripped so
+/// a route switch never leaks Google-only fields to a foreign gateway.
+fn strip_google_tool_call_extra_content(messages: &mut [Value]) {
+    for message in messages {
+        let Some(tool_calls) = message.get_mut("tool_calls").and_then(Value::as_array_mut) else {
+            continue;
+        };
+        for call in tool_calls {
+            if let Some(extra) = call.get_mut("extra_content") {
+                if let Some(obj) = extra.as_object_mut() {
+                    obj.remove("google");
+                    if obj.is_empty() {
+                        call.as_object_mut().map(|c| c.remove("extra_content"));
+                    }
+                }
+            }
+        }
+    }
+}
+
 fn mistral_model_has_adjustable_reasoning(model: &str) -> bool {
     let model = model.trim().to_ascii_lowercase();
     model.starts_with("mistral-medium") || model.starts_with("mistral-small")
@@ -870,6 +989,7 @@ pub(crate) fn build_chat_wire_body(
         let wire = wire_model_for_provider_route(provider, base_url, &request.model);
         crate::models::effective_muse_wire_id(&wire).to_string()
     };
+    validate_google_thought_signature_replay(provider, base_url, &model, &messages)?;
     let mut body = if stream {
         json!({
             "model": model.clone(),
@@ -1468,6 +1588,9 @@ impl<'a> PromptBuilder<'a> {
         if is_exact_mistral_chat_route(provider, base_url) {
             reshape_mistral_messages_for_reasoning_replay(&mut messages);
         }
+        if !is_exact_google_chat_route(provider, base_url) {
+            strip_google_tool_call_extra_content(&mut messages);
+        }
         messages
     }
 
@@ -2328,7 +2451,7 @@ fn build_chat_messages_with_reasoning(
                     name,
                     input,
                     caller,
-                    ..
+                    thought_signature,
                 } => {
                     let args = serde_json::to_string(input).unwrap_or_else(|_| input.to_string());
                     let mut call = json!({
@@ -2339,6 +2462,9 @@ fn build_chat_messages_with_reasoning(
                             "arguments": args,
                         }
                     });
+                    if let Some(signature) = thought_signature {
+                        call["extra_content"]["google"]["thought_signature"] = json!(signature);
+                    }
                     if let Some(caller) = caller {
                         call["caller"] = json!({
                             "type": caller.caller_type,
@@ -3317,11 +3443,16 @@ fn parse_chat_message_for_route(
                     })
             });
 
+            let thought_signature = call
+                .pointer("/extra_content/google/thought_signature")
+                .and_then(Value::as_str)
+                .map(str::to_string);
             content_blocks.push(ContentBlock::ToolUse {
                 id,
                 name: from_api_tool_name(&name),
                 input: arguments,
                 caller,
+                thought_signature,
             });
         }
     }
@@ -3771,6 +3902,10 @@ fn parse_sse_chunk_with_reasoning_style(
                                 })
                             });
 
+                            let thought_signature = tc
+                                .pointer("/extra_content/google/thought_signature")
+                                .and_then(Value::as_str)
+                                .map(str::to_string);
                             events.push(StreamEvent::ContentBlockStart {
                                 index: block_index,
                                 content_block: ContentBlockStart::ToolUse {
@@ -3778,6 +3913,7 @@ fn parse_sse_chunk_with_reasoning_style(
                                     name: from_api_tool_name(&name),
                                     input: json!({}),
                                     caller,
+                                    thought_signature,
                                 },
                             });
                             *content_index = (*content_index).saturating_add(1);
@@ -4201,6 +4337,7 @@ mod minimax_reasoning_replay_tests {
                         name: "read".to_string(),
                         input: serde_json::json!({ "path": "widget.rs" }),
                         caller: None,
+                        thought_signature: None,
                     },
                 ],
             },
@@ -5803,6 +5940,7 @@ mod image_block_wire_tests {
                     name: "read".to_string(),
                     input: serde_json::json!({"path": "shot.png"}),
                     caller: None,
+                    thought_signature: None,
                 }],
             },
             Message {
@@ -5874,6 +6012,7 @@ mod mistral_reasoning_tests {
                             name: "read_file".to_string(),
                             input: json!({"path": "README.md"}),
                             caller: None,
+                            thought_signature: None,
                         },
                     ],
                 },
@@ -6258,3 +6397,270 @@ mod mistral_reasoning_tests {
         }
     }
 }
+
+#[cfg(test)]
+mod google_thought_signature_tests {
+    use super::*;
+
+    // ── Google thought signatures (#v0.9.8 Google backend) ──────────────
+    use crate::config::{DEFAULT_GOOGLE_BASE_URL, DEFAULT_OPENAI_BASE_URL};
+
+    fn google_request_with_signed_tool(signature: Option<&str>) -> MessageRequest {
+        MessageRequest {
+            model: "gemini-3.1-pro-preview".to_string(),
+            messages: vec![
+                Message {
+                    role: "user".to_string(),
+                    content: vec![ContentBlock::Text {
+                        text: "Read the config.".to_string(),
+                        cache_control: None,
+                    }],
+                },
+                Message {
+                    role: "assistant".to_string(),
+                    content: vec![
+                        ContentBlock::Text {
+                            text: "Reading now.".to_string(),
+                            cache_control: None,
+                        },
+                        ContentBlock::ToolUse {
+                            id: "call-g-1".to_string(),
+                            name: "read".to_string(),
+                            input: json!({"path": "config.toml"}),
+                            caller: None,
+                            thought_signature: signature.map(str::to_string),
+                        },
+                    ],
+                },
+                Message {
+                    role: "user".to_string(),
+                    content: vec![ContentBlock::ToolResult {
+                        tool_use_id: "call-g-1".to_string(),
+                        content: "key = \"value\"".to_string(),
+                        is_error: None,
+                        content_blocks: None,
+                    }],
+                },
+            ],
+            max_tokens: 64,
+            system: None,
+            tools: None,
+            tool_choice: None,
+            metadata: None,
+            thinking: None,
+            reasoning_effort: Some("high".to_string()),
+            stream: None,
+            temperature: None,
+            top_p: None,
+        }
+    }
+
+    #[test]
+    fn google_route_round_trips_thought_signatures_on_replayed_tool_calls() {
+        let request = google_request_with_signed_tool(Some("SIG-abc123"));
+        let messages = build_chat_messages_for_request_and_provider_and_route(
+            &request,
+            ApiProvider::Google,
+            DEFAULT_GOOGLE_BASE_URL,
+        );
+        let assistant = messages
+            .iter()
+            .find(|m| m.get("role") == Some(&json!("assistant")))
+            .expect("assistant replay message");
+        let signature = assistant
+            .pointer("/tool_calls/0/extra_content/google/thought_signature")
+            .and_then(serde_json::Value::as_str);
+        assert_eq!(signature, Some("SIG-abc123"));
+    }
+
+    #[test]
+    fn google_route_fails_closed_when_replayed_signature_is_missing() {
+        let request = google_request_with_signed_tool(None);
+        let error = build_chat_wire_body(
+            &request,
+            ApiProvider::Google,
+            DEFAULT_GOOGLE_BASE_URL,
+            false,
+        )
+        .err()
+        .expect("missing signature must fail closed before transport");
+        assert!(
+            error.to_string().contains("thought signature"),
+            "error must name the missing signature: {error}"
+        );
+    }
+
+    #[test]
+    fn google_missing_signature_is_a_warning_not_an_error_for_flash_lite() {
+        // 2.5 Flash-Lite ships thinking off; Google may legitimately omit
+        // signatures there, so replay proceeds.
+        let mut request = google_request_with_signed_tool(None);
+        request.model = "gemini-2.5-flash-lite".to_string();
+        build_chat_wire_body(
+            &request,
+            ApiProvider::Google,
+            DEFAULT_GOOGLE_BASE_URL,
+            false,
+        )
+        .expect("flash-lite replay must not require a signature");
+    }
+
+    #[test]
+    fn non_google_routes_never_see_google_extra_content() {
+        let request = google_request_with_signed_tool(Some("SIG-abc123"));
+        let messages = build_chat_messages_for_request_and_provider_and_route(
+            &request,
+            ApiProvider::Openai,
+            DEFAULT_OPENAI_BASE_URL,
+        );
+        for message in &messages {
+            if let Some(tool_calls) = message.get("tool_calls").and_then(|v| v.as_array()) {
+                for call in tool_calls {
+                    assert!(
+                        call.get("extra_content").is_none(),
+                        "Google-only fields must not leak to other providers"
+                    );
+                }
+            }
+        }
+    }
+
+    #[test]
+    fn google_neighbor_base_url_does_not_get_google_dialect() {
+        // A Google provider row pointed at some other gateway must not
+        // carry signatures or fail closed: the dialect binds to the exact
+        // official route, not to provider identity alone.
+        let request = google_request_with_signed_tool(None);
+        build_chat_wire_body(
+            &request,
+            ApiProvider::Google,
+            "https://gateway.example.com/v1",
+            false,
+        )
+        .expect("non-official Google base URL must not require signatures");
+        let messages = build_chat_messages_for_request_and_provider_and_route(
+            &google_request_with_signed_tool(Some("SIG")),
+            ApiProvider::Google,
+            "https://gateway.example.com/v1",
+        );
+        assert!(
+            messages
+                .iter()
+                .all(|m| m.pointer("/tool_calls/0/extra_content").is_none()),
+            "signatures must not be sent to a non-Google endpoint"
+        );
+    }
+
+    #[test]
+    fn google_thinking_level_maps_effort_onto_documented_body_field() {
+        let mut request = google_request_with_signed_tool(Some("SIG"));
+        request.reasoning_effort = Some("high".to_string());
+        let body = build_chat_wire_body(
+            &request,
+            ApiProvider::Google,
+            DEFAULT_GOOGLE_BASE_URL,
+            false,
+        )
+        .expect("valid google body");
+        assert_eq!(
+            body.body
+                .pointer("/google/thinking_config/thinking_level")
+                .and_then(serde_json::Value::as_str),
+            Some("high")
+        );
+
+        let mut low = google_request_with_signed_tool(Some("SIG"));
+        low.reasoning_effort = Some("low".to_string());
+        let body = build_chat_wire_body(&low, ApiProvider::Google, DEFAULT_GOOGLE_BASE_URL, false)
+            .expect("valid google body");
+        assert_eq!(
+            body.body
+                .pointer("/google/thinking_config/thinking_level")
+                .and_then(serde_json::Value::as_str),
+            Some("low")
+        );
+    }
+
+    #[test]
+    fn google_signature_captured_from_non_streaming_tool_call() {
+        let payload = json!({
+            "id": "resp-1",
+            "model": "gemini-3.1-pro-preview",
+            "choices": [{
+                "index": 0,
+                "finish_reason": "tool_calls",
+                "message": {
+                    "role": "assistant",
+                    "content": null,
+                    "tool_calls": [{
+                        "id": "call-g-9",
+                        "type": "function",
+                        "function": {
+                            "name": "read",
+                            "arguments": "{\"path\":\"x\"}"
+                        },
+                        "extra_content": {
+                            "google": { "thought_signature": "SIG-stream" }
+                        }
+                    }]
+                }
+            }],
+            "usage": {"prompt_tokens": 1, "completion_tokens": 1, "total_tokens": 2}
+        });
+        let response = parse_chat_message(&payload).expect("parses");
+        let signature = response.content.iter().find_map(|block| match block {
+            ContentBlock::ToolUse {
+                thought_signature, ..
+            } => thought_signature.clone(),
+            _ => None,
+        });
+        assert_eq!(signature.as_deref(), Some("SIG-stream"));
+    }
+
+    #[test]
+    fn google_signature_captured_from_streaming_first_chunk() {
+        let chunk = json!({
+            "choices": [{
+                "index": 0,
+                "delta": {
+                    "tool_calls": [{
+                        "index": 0,
+                        "id": "call-g-7",
+                        "type": "function",
+                        "function": { "name": "read", "arguments": "{}" },
+                        "extra_content": {
+                            "google": { "thought_signature": "SIG-delta" }
+                        }
+                    }]
+                }
+            }]
+        });
+        let mut content_index = 0u32;
+        let mut text_started = false;
+        let mut thinking_started = false;
+        let mut tool_indices = std::collections::HashMap::new();
+        let mut reasoning_buffers = std::collections::HashMap::new();
+        let mut inline_tags = InlineReasoningTagState::default();
+        let events = parse_sse_chunk_with_reasoning_style(
+            &chunk,
+            &mut content_index,
+            &mut text_started,
+            &mut thinking_started,
+            &mut tool_indices,
+            &mut reasoning_buffers,
+            &mut inline_tags,
+            ReasoningStreamStyle::None,
+        );
+        let signature = events.iter().find_map(|event| match event {
+            StreamEvent::ContentBlockStart {
+                content_block:
+                    ContentBlockStart::ToolUse {
+                        thought_signature, ..
+                    },
+                ..
+            } => thought_signature.clone(),
+            _ => None,
+        });
+        assert_eq!(signature.as_deref(), Some("SIG-delta"));
+    }
+}

crates/tui/src/client/chat/tests/stream_decoder.rs

@@ -634,7 +634,7 @@ fn decoder_emits_tool_use_block_for_tool_call_delta() {
         events.iter().any(|e| matches!(
             e,
             StreamEvent::ContentBlockStart {
-                content_block: ContentBlockStart::ToolUse { name, .. },
+                content_block: ContentBlockStart::ToolUse { name, ..},
                 ..
             } if name == "grep_files"
         )),
@@ -662,7 +662,7 @@ fn decoder_uses_fallback_name_for_empty_streaming_tool_name() {
         events.iter().any(|event| matches!(
             event,
             StreamEvent::ContentBlockStart {
-                content_block: ContentBlockStart::ToolUse { name, .. },
+                content_block: ContentBlockStart::ToolUse { name, ..},
                 ..
             } if name == "unknown_tool"
         )),
@@ -784,6 +784,7 @@ fn tool_use_message(id: &str, name: &str, input: Value) -> Message {
             name: name.to_string(),
             input,
             caller: None,
+            thought_signature: None,
         }],
     }
 }

crates/tui/src/client/ds4_tests.rs

@@ -121,7 +121,7 @@ fn non_streaming_fixture_preserves_tool_call_and_usage() -> Result<()> {
 
     assert!(matches!(
         response.content.as_slice(),
-        [ContentBlock::ToolUse { id, name, input, .. }]
+        [ContentBlock::ToolUse { id, name, input, ..}]
             if id == "call_ds4_0"
                 && name == "read_file"
                 && input == &json!({"path": "src/main.rs"})
@@ -152,7 +152,7 @@ fn malformed_tool_arguments_remain_visible_for_feedback() -> Result<()> {
 
     assert!(matches!(
         response.content.as_slice(),
-        [ContentBlock::ToolUse { input: Value::String(raw), .. }] if raw == "{bad json"
+        [ContentBlock::ToolUse { input: Value::String(raw), ..}] if raw == "{bad json"
     ));
     Ok(())
 }

crates/tui/src/client/responses.rs

@@ -333,7 +333,8 @@ impl DeepSeekClient {
                                                         name: from_api_tool_name(&name),
                                                         input: json!({}),
                                                         caller: None,
-                                                    },
+                                                    thought_signature: None,
+                                                },
                                             });
                                             current_block_index =
                                                 Some(content_block_counter - 1);
@@ -562,11 +563,13 @@ impl DeepSeekClient {
                             name,
                             input,
                             caller,
+                            thought_signature,
                         } => ContentBlock::ToolUse {
                             id,
                             name,
                             input,
                             caller,
+                            thought_signature,
                         },
                         ContentBlockStart::ServerToolUse { id, name, input } => {
                             ContentBlock::ServerToolUse { id, name, input }

crates/tui/src/client/responses/tests.rs

@@ -938,6 +938,7 @@ fn responses_input_includes_user_role_tool_results() {
                     name: "checklist_write".to_string(),
                     input: json!({"items": []}),
                     caller: None,
+                    thought_signature: None,
                 }],
             },
             Message {
@@ -983,6 +984,7 @@ fn responses_input_encodes_tool_call_names() {
                 name: "web.run".to_string(),
                 input: json!({}),
                 caller: None,
+                thought_signature: None,
             }],
         }],
         max_tokens: 128,
@@ -1158,6 +1160,7 @@ fn tool_result_image_becomes_native_function_output_content() {
                 name: "read".to_string(),
                 input: serde_json::json!({"path": "shot.png"}),
                 caller: None,
+                thought_signature: None,
             }],
         },
         Message {

crates/tui/src/commands/groups/debug/tests.rs

@@ -733,6 +733,7 @@ fn cache_inspect_displays_tool_result_budget_metadata() {
             name: "shell_command".to_string(),
             input: serde_json::json!({"command": "cargo test"}),
             caller: None,
+            thought_signature: None,
         }],
     });
     app.api_messages.push(Message {
@@ -751,6 +752,7 @@ fn cache_inspect_displays_tool_result_budget_metadata() {
             name: "shell_command".to_string(),
             input: serde_json::json!({"command": "cargo test"}),
             caller: None,
+            thought_signature: None,
         }],
     });
     app.api_messages.push(Message {
@@ -1466,6 +1468,7 @@ fn test_patch_undo_prunes_tool_turn_context() {
                 name: "write_file".to_string(),
                 input: serde_json::json!({"path": "a.txt"}),
                 caller: None,
+                thought_signature: None,
             },
         ],
     });
@@ -1648,12 +1651,14 @@ fn test_prune_undone_tool_context_preserves_prior_tool_pairs() {
                 name: "write_file".to_string(),
                 input: serde_json::json!({"path": "a.txt"}),
                 caller: None,
+                thought_signature: None,
             },
             ContentBlock::ToolUse {
                 id: "call-b".to_string(),
                 name: "write_file".to_string(),
                 input: serde_json::json!({"path": "b.txt"}),
                 caller: None,
+                thought_signature: None,
             },
         ],
     });
@@ -1691,7 +1696,7 @@ fn test_prune_undone_tool_context_preserves_prior_tool_pairs() {
         &app.api_messages[1].content[..],
         [
             ContentBlock::Text { .. },
-            ContentBlock::ToolUse { id, .. }
+            ContentBlock::ToolUse { id, ..}
         ] if id == "call-a"
     ));
     assert!(matches!(

crates/tui/src/commands/groups/debug/undo.rs

@@ -61,7 +61,7 @@ pub(crate) fn prune_undone_tool_context(app: &mut App, tool_id: &str) {
                 msg.content
                     .iter()
                     .position(
-                        |block| matches!(block, ContentBlock::ToolUse { id, .. } if id == tool_id),
+                        |block| matches!(block, ContentBlock::ToolUse { id, ..} if id == tool_id),
                     )
                     .map(|block_idx| (msg_idx, block_idx))
             })

crates/tui/src/commands/groups/session/export.rs

@@ -476,6 +476,7 @@ fn render_content_block(out: &mut String, index: usize, block: &ContentBlock) {
             name,
             input,
             caller,
+            ..
         } => {
             let _ = writeln!(out, "### Content {index}: Tool call\n");
             let _ = writeln!(out, "- ID: {}", inline_text(id));
@@ -961,6 +962,7 @@ mod tests {
                             caller_type: "code_execution_20250825".to_string(),
                             tool_id: Some("server-tool-1".to_string()),
                         }),
+                        thought_signature: None,
                     },
                 ],
             },

crates/tui/src/commands/groups/session/structcopy.rs

@@ -278,6 +278,7 @@ fn block_payload(block: &ContentBlock) -> Value {
             name,
             input,
             caller,
+            ..
         } => json!({
             "type": "tool_use",
             "id": id,
@@ -1294,6 +1295,7 @@ mod tests {
                             caller_type: "code_execution_20250825".to_string(),
                             tool_id: None,
                         }),
+                        thought_signature: None,
                     },
                 ],
             },
@@ -1428,6 +1430,7 @@ mod tests {
             name: "view_image".to_string(),
             input: json!({}),
             caller: None,
+            thought_signature: None,
         });
         let json = stdout_json(&execute_structcopy(
             &mut app,
@@ -1456,6 +1459,7 @@ mod tests {
                     input: json!({}),
                     // No caller recorded: also an unknown, also null.
                     caller: None,
+                    thought_signature: None,
                 }],
             },
             Message {
@@ -1740,6 +1744,7 @@ mod tests {
             name: "exec_command".to_string(),
             input: json!({}),
             caller: None,
+            thought_signature: None,
         });
         let json = stdout_json(&execute_structcopy(
             &mut app,
@@ -1765,6 +1770,7 @@ mod tests {
                 name: "exec_command".to_string(),
                 input: json!({}),
                 caller: None,
+                thought_signature: None,
             });
             let json = stdout_json(&execute_structcopy(
                 &mut app,
@@ -1804,6 +1810,7 @@ mod tests {
                 name: "exec_command".to_string(),
                 input,
                 caller: None,
+                thought_signature: None,
             }],
         }];
 
@@ -1875,6 +1882,7 @@ mod tests {
                     "pass\u{7}word": "another-plain-value-that-must-not-leak",
                 }),
                 caller: None,
+                thought_signature: None,
             }],
         }];
 
@@ -1940,6 +1948,7 @@ mod tests {
                 name: "exec_command".to_string(),
                 input: json!({exact: 1, same_after_flatten: 2}),
                 caller: None,
+                thought_signature: None,
             }],
         }];
 
@@ -2254,6 +2263,7 @@ mod tests {
                     "unc": r"\\server\private\customer.txt",
                 }),
                 caller: None,
+                thought_signature: None,
             }],
         }];
 
@@ -2411,6 +2421,7 @@ mod tests {
                 name: "exec_command".to_string(),
                 input: json!({"a": {"b": {"c": {"d": {"e": "too deep"}}}}}),
                 caller: None,
+                thought_signature: None,
             }],
         }];
         let caps = Caps {
@@ -2453,6 +2464,7 @@ mod tests {
                     "deep": {"one": {"two": ["cut-a", "cut-b"]}},
                 }),
                 caller: None,
+                thought_signature: None,
             }],
         }];
         let caps = Caps {
@@ -2521,6 +2533,7 @@ mod tests {
                 name: "exec_command".to_string(),
                 input: json!({"deep": {"one": {long_a: 1, long_b: 2}}}),
                 caller: None,
+                thought_signature: None,
             }],
         }];
         let caps = Caps {

crates/tui/src/compaction.rs

@@ -1307,6 +1307,7 @@ mod tests {
                 name: name.to_string(),
                 input,
                 caller: None,
+                thought_signature: None,
             }],
         }
     }
@@ -1841,6 +1842,7 @@ mod tests {
                         name: "read_file".to_string(),
                         input: serde_json::json!({"path": "Cargo.toml"}),
                         caller: None,
+                        thought_signature: None,
                     },
                 ],
             },

crates/tui/src/compaction/tests.rs

@@ -192,6 +192,7 @@ fn oversized_tool_pair(id: &str, content: String) -> Vec<Message> {
                 name: "read_file".to_string(),
                 input: serde_json::json!({"path": "src/compaction.rs"}),
                 caller: None,
+                thought_signature: None,
             }],
         },
         Message {

crates/tui/src/config.rs

@@ -88,6 +88,10 @@ pub enum ApiProvider {
     Xai,
     /// Mistral AI — la Plateforme (OpenAI-compatible Chat Completions).
     Mistral,
+    /// Google Gemini — official OpenAI-compatible endpoint. A distinct
+    /// backend, not an OpenAI alias: thought signatures on tool calls are
+    /// captured and replayed per Google's contract.
+    Google,
     /// Jiangsu Telecom TokenHub — OpenAI-compatible AI gateway.
     Telecomjs,
     /// Alibaba Cloud Model Studio — Token Plan (OpenAI-compatible Chat Completions).
@@ -317,7 +321,7 @@ impl ApiProvider {
 
     /// `ApiProvider` discriminant → `ProviderKind` lookup.
     /// Index 1 is `None` for the legacy `DeepseekCN` variant.
-    const KIND_LOOKUP: [Option<codewhale_config::ProviderKind>; 44] = [
+    const KIND_LOOKUP: [Option<codewhale_config::ProviderKind>; 45] = [
         Some(codewhale_config::ProviderKind::Deepseek),
         None, // DeepseekCN
         Some(codewhale_config::ProviderKind::DeepseekAnthropic),
@@ -356,6 +360,7 @@ impl ApiProvider {
         Some(codewhale_config::ProviderKind::Meta),
         Some(codewhale_config::ProviderKind::Xai),
         Some(codewhale_config::ProviderKind::Mistral),
+        Some(codewhale_config::ProviderKind::Google),
         Some(codewhale_config::ProviderKind::Telecomjs),
         Some(codewhale_config::ProviderKind::ModelstudioTokenPlan),
         Some(codewhale_config::ProviderKind::ModelstudioTokenPlanAnthropic),
@@ -365,7 +370,7 @@ impl ApiProvider {
     ];
 
     /// `ProviderKind` discriminant → `ApiProvider` lookup.
-    const FROM_KIND_LOOKUP: [Self; 43] = [
+    const FROM_KIND_LOOKUP: [Self; 44] = [
         Self::Deepseek,
         Self::DeepseekAnthropic,
         Self::NvidiaNim,
@@ -408,6 +413,7 @@ impl ApiProvider {
         Self::ModelstudioTokenPlanAnthropic,
         Self::ModelstudioCodingPlan,
         Self::ModelstudioCodingPlanAnthropic,
+        Self::Google,
         Self::Custom,
     ];
 
@@ -1549,6 +1555,15 @@ pub fn model_completion_names_for_provider(provider: ApiProvider) -> Vec<&'stati
             "mistral-small-latest",
             "mistral-large-latest",
         ],
+        ApiProvider::Google => vec![
+            DEFAULT_GOOGLE_MODEL,
+            "gemini-3-pro-preview",
+            "gemini-3.6-flash",
+            "gemini-3.5-flash",
+            "gemini-3.5-flash-lite",
+            "gemini-2.5-pro",
+            "gemini-2.5-flash",
+        ],
         // Custom endpoints expose no built-in completion names; the user
         // supplies their own model id (#1519).
         ApiProvider::Custom => Vec::new(),
@@ -3442,6 +3457,13 @@ pub struct ProvidersConfig {
         alias = "la_plateforme"
     )]
     pub mistral: ProviderConfig,
+    #[serde(
+        default,
+        alias = "google-gemini",
+        alias = "google_gemini",
+        alias = "gemini"
+    )]
+    pub google: ProviderConfig,
     #[serde(
         default,
         alias = "telecom-js",
@@ -4899,6 +4921,7 @@ impl Config {
             ApiProvider::Meta => &providers.meta,
             ApiProvider::Xai => &providers.xai,
             ApiProvider::Mistral => &providers.mistral,
+            ApiProvider::Google => &providers.google,
             ApiProvider::Telecomjs => &providers.telecomjs,
             ApiProvider::ModelstudioTokenPlan => &providers.modelstudio_token_plan,
             ApiProvider::ModelstudioTokenPlanAnthropic => {
@@ -4976,6 +4999,7 @@ impl Config {
             ApiProvider::Meta => &mut providers.meta,
             ApiProvider::Xai => &mut providers.xai,
             ApiProvider::Mistral => &mut providers.mistral,
+            ApiProvider::Google => &mut providers.google,
             ApiProvider::Telecomjs => &mut providers.telecomjs,
             ApiProvider::ModelstudioTokenPlan => &mut providers.modelstudio_token_plan,
             ApiProvider::ModelstudioTokenPlanAnthropic => {
@@ -5338,6 +5362,7 @@ impl Config {
             ApiProvider::Meta => DEFAULT_META_MODEL,
             ApiProvider::Xai => DEFAULT_XAI_MODEL,
             ApiProvider::Mistral => DEFAULT_MISTRAL_MODEL,
+            ApiProvider::Google => DEFAULT_GOOGLE_MODEL,
             ApiProvider::Telecomjs => DEFAULT_TELECOMJS_MODEL,
             ApiProvider::ModelstudioTokenPlan
             | ApiProvider::ModelstudioTokenPlanAnthropic
@@ -5457,6 +5482,7 @@ impl Config {
             | ApiProvider::Meta
             | ApiProvider::Xai
             | ApiProvider::Mistral
+            | ApiProvider::Google
             | ApiProvider::Telecomjs
             | ApiProvider::ModelstudioTokenPlan
             | ApiProvider::ModelstudioTokenPlanAnthropic
@@ -5563,6 +5589,7 @@ impl Config {
                         ApiProvider::Meta => DEFAULT_META_BASE_URL,
                         ApiProvider::Xai => DEFAULT_XAI_BASE_URL,
                         ApiProvider::Mistral => DEFAULT_MISTRAL_BASE_URL,
+                        ApiProvider::Google => DEFAULT_GOOGLE_BASE_URL,
                         ApiProvider::Telecomjs => DEFAULT_TELECOMJS_BASE_URL,
                         ApiProvider::ModelstudioTokenPlan
                         | ApiProvider::ModelstudioTokenPlanAnthropic
@@ -7133,6 +7160,7 @@ fn provider_env_base_url_override(provider: ApiProvider) -> Option<String> {
         ApiProvider::Meta => &["META_MODEL_API_BASE_URL", "MODEL_API_BASE_URL"],
         ApiProvider::Xai => &["XAI_BASE_URL"],
         ApiProvider::Mistral => &["MISTRAL_BASE_URL"],
+        ApiProvider::Google => &["GOOGLE_BASE_URL", "GEMINI_BASE_URL"],
         ApiProvider::Telecomjs => &["TELECOMJS_BASE_URL"],
         ApiProvider::ModelstudioTokenPlan | ApiProvider::ModelstudioTokenPlanAnthropic => {
             &["MODELSTUDIO_TOKEN_PLAN_BASE_URL"]
@@ -7477,6 +7505,13 @@ fn apply_env_overrides_unlocked(config: &mut Config, policy: ConfigEnvironmentPo
                     .mistral
                     .base_url = Some(value);
             }
+            ApiProvider::Google => {
+                config
+                    .providers
+                    .get_or_insert_with(ProvidersConfig::default)
+                    .google
+                    .base_url = Some(value);
+            }
             ApiProvider::Telecomjs => {
                 config
                     .providers
@@ -7838,6 +7873,7 @@ fn apply_env_overrides_unlocked(config: &mut Config, policy: ConfigEnvironmentPo
                 ApiProvider::Meta => &mut providers.meta,
                 ApiProvider::Xai => &mut providers.xai,
                 ApiProvider::Mistral => &mut providers.mistral,
+                ApiProvider::Google => &mut providers.google,
                 ApiProvider::Telecomjs => &mut providers.telecomjs,
                 ApiProvider::ModelstudioTokenPlan => &mut providers.modelstudio_token_plan,
                 ApiProvider::ModelstudioTokenPlanAnthropic => {
@@ -8183,6 +8219,7 @@ fn apply_env_overrides_unlocked(config: &mut Config, policy: ConfigEnvironmentPo
                 ApiProvider::Meta => &mut providers.meta,
                 ApiProvider::Xai => &mut providers.xai,
                 ApiProvider::Mistral => &mut providers.mistral,
+                ApiProvider::Google => &mut providers.google,
                 ApiProvider::Telecomjs => &mut providers.telecomjs,
                 ApiProvider::ModelstudioTokenPlan => &mut providers.modelstudio_token_plan,
                 ApiProvider::ModelstudioTokenPlanAnthropic => {
@@ -9581,6 +9618,7 @@ fn merge_providers(
             meta: merge_provider_config(base.meta, override_cfg.meta),
             xai: merge_provider_config(base.xai, override_cfg.xai),
             mistral: merge_provider_config(base.mistral, override_cfg.mistral),
+            google: merge_provider_config(base.google, override_cfg.google),
             telecomjs: merge_provider_config(base.telecomjs, override_cfg.telecomjs),
             modelstudio_token_plan: merge_provider_config(
                 base.modelstudio_token_plan,

crates/tui/src/config/models.rs

@@ -222,6 +222,9 @@ pub const XAI_GROK_4_20_0309_NON_REASONING_MODEL: &str = "grok-4.20-0309-non-rea
 pub const DEFAULT_XAI_BASE_URL: &str = "https://api.x.ai/v1";
 pub const DEFAULT_MISTRAL_MODEL: &str = "mistral-code-latest";
 pub const DEFAULT_MISTRAL_BASE_URL: &str = "https://api.mistral.ai/v1";
+pub const DEFAULT_GOOGLE_MODEL: &str = "gemini-3.1-pro-preview";
+pub const DEFAULT_GOOGLE_BASE_URL: &str =
+    "https://generativelanguage.googleapis.com/v1beta/openai/";
 pub const DEFAULT_TELECOMJS_MODEL: &str = "deepseek-v4-pro";
 pub const DEFAULT_TELECOMJS_BASE_URL: &str = "https://aigw.telecomjs.com/v1";
 // Alibaba Cloud Model Studio (DashScope) defaults

crates/tui/src/config_persistence.rs

@@ -314,6 +314,7 @@ fn provider_base_url_table_key(provider: ApiProvider) -> anyhow::Result<&'static
         ApiProvider::Meta => Ok("meta"),
         ApiProvider::Xai => Ok("xai"),
         ApiProvider::Mistral => Ok("mistral"),
+        ApiProvider::Google => Ok("google"),
         ApiProvider::Telecomjs => Ok("telecomjs"),
         ApiProvider::ModelstudioTokenPlan => Ok("modelstudio_token_plan"),
         ApiProvider::ModelstudioTokenPlanAnthropic => Ok("modelstudio_token_plan_anthropic"),

crates/tui/src/core/engine/streaming.rs

@@ -20,6 +20,9 @@ pub(super) struct ToolUseState {
     pub(super) name: String,
     pub(super) input: serde_json::Value,
     pub(super) caller: Option<ToolCaller>,
+    /// Google thought signature captured on the tool call; replayed with the
+    /// assistant tool-call message on later turns.
+    pub(super) thought_signature: Option<String>,
     pub(super) input_buffer: String,
     pub(super) input_parse_error: Option<String>,
 }

crates/tui/src/core/engine/tests.rs

@@ -11696,6 +11696,7 @@ async fn session_update_preserves_reasoning_tool_only_turn() {
                 name: "read_file".to_string(),
                 input: json!({"path": "Cargo.toml"}),
                 caller: None,
+                thought_signature: None,
             },
         ],
     };
@@ -14668,6 +14669,7 @@ fn turn_metadata_skips_tool_result_messages() {
             name: "read_file".to_string(),
             input: serde_json::json!({"path": "src/lib.rs"}),
             caller: None,
+            thought_signature: None,
         }],
     });
     // Tool result, stored as role="user" internally.
@@ -15665,6 +15667,7 @@ fn tool_state(initial: serde_json::Value, buffer: &str) -> ToolUseState {
         name: "exec_shell".into(),
         input: initial,
         caller: None,
+        thought_signature: None,
         input_buffer: buffer.into(),
         input_parse_error: None,
     }

crates/tui/src/core/engine/turn_loop.rs

@@ -1315,6 +1315,7 @@ impl Engine {
                             name,
                             input,
                             caller,
+                            thought_signature,
                         } => {
                             crate::logging::info(format!(
                                 "Tool '{name}' block start. Initial input: {input:?}"
@@ -1330,6 +1331,7 @@ impl Engine {
                                 name,
                                 input,
                                 caller,
+                                thought_signature,
                                 input_buffer: String::new(),
                                 input_parse_error: None,
                             });
@@ -1345,6 +1347,7 @@ impl Engine {
                                 name,
                                 input,
                                 caller: None,
+                                thought_signature: None,
                                 input_buffer: String::new(),
                                 input_parse_error: None,
                             });
@@ -1691,6 +1694,7 @@ impl Engine {
                                 name: tool.name.clone(),
                                 input: tool.input.clone(),
                                 caller: tool.caller.clone(),
+                                thought_signature: tool.thought_signature.clone(),
                             });
                         }
                         let has_sendable_assistant_content = resume_blocks.iter().any(|block| {
@@ -1768,6 +1772,7 @@ impl Engine {
                         name: call.name,
                         input: call.args,
                         caller: None,
+                        thought_signature: None,
                         input_buffer: String::new(),
                         input_parse_error: None,
                     });
@@ -1797,6 +1802,7 @@ impl Engine {
                     name: tool.name.clone(),
                     input: tool.input.clone(),
                     caller: tool.caller.clone(),
+                    thought_signature: tool.thought_signature.clone(),
                 });
             }
 
@@ -5094,12 +5100,12 @@ mod tests {
     /// assistant message, the streaming parser emits the events in this order:
     ///
     /// ```text
-    /// ContentBlockStart::ToolUse { index: 0, .. }   // tool #1
+    /// ContentBlockStart::ToolUse { index: 0, ..}   // tool #1
     /// ContentBlockDelta { index: 0, .. }            // its arguments
-    /// ContentBlockStart::ToolUse { index: 1, .. }   // tool #2
+    /// ContentBlockStart::ToolUse { index: 1, ..}   // tool #2
     /// ContentBlockDelta { index: 1, .. }
     /// …
-    /// ContentBlockStart::ToolUse { index: N-1, .. }
+    /// ContentBlockStart::ToolUse { index: N-1, ..}
     /// ContentBlockDelta { index: N-1, .. }
     /// ContentBlockStop { index: 0 }                 // ── only flushed at
     /// ContentBlockStop { index: 1 }                 //    finish_reason
@@ -5119,7 +5125,7 @@ mod tests {
         let mut current_tool_indices: std::collections::HashMap<u32, usize> =
             std::collections::HashMap::new();
 
-        // Simulate `ContentBlockStart::ToolUse { index: i }` for 7 tools.
+        // Simulate `ContentBlockStart::ToolUse { index: i, ..}` for 7 tools.
         for block_index in 0..7u32 {
             current_tool_indices.insert(block_index, block_index as usize);
         }

crates/tui/src/lib.rs

@@ -15760,6 +15760,7 @@ mod terminal_mode_tests {
                         name: "exec_shell".to_string(),
                         input: serde_json::json!({"command": "cargo test"}),
                         caller: None,
+                        thought_signature: None,
                     },
                 ],
             },

crates/tui/src/llm_client/mock.rs

@@ -400,6 +400,7 @@ pub mod canned {
                 name: name.to_string(),
                 input: Value::Null,
                 caller: None,
+                thought_signature: None,
             },
         }
     }

crates/tui/src/model_registry.rs

@@ -75,6 +75,8 @@ pub enum ModelProvider {
     Xai,
     /// Mistral AI la Plateforme models.
     Mistral,
+    /// Google Gemini models (official OpenAI-compatible route).
+    Google,
     /// Anything not otherwise classified (still gets real metadata via the
     /// `models.rs` heuristics where possible).
     Other,
@@ -201,6 +203,15 @@ const SEED_MODEL_IDS: &[(&str, ModelProvider)] = &[
     ("grok-4.20-0309-non-reasoning", ModelProvider::Xai),
     // --- Mistral AI (current first-party roster; deprecated Magistral remains
     // accepted through the long-tail models.rs compatibility path) ---
+    // --- Google Gemini (official OpenAI-compatible route; preview flagships
+    // as listed by Google's model pages, 2026-08).
+    ("gemini-3.1-pro-preview", ModelProvider::Google),
+    ("gemini-3-pro-preview", ModelProvider::Google),
+    ("gemini-3.6-flash", ModelProvider::Google),
+    ("gemini-3.5-flash", ModelProvider::Google),
+    ("gemini-3.5-flash-lite", ModelProvider::Google),
+    ("gemini-2.5-pro", ModelProvider::Google),
+    ("gemini-2.5-flash", ModelProvider::Google),
     ("mistral-code-latest", ModelProvider::Mistral),
     ("mistral-medium-latest", ModelProvider::Mistral),
     ("mistral-small-latest", ModelProvider::Mistral),

crates/tui/src/models.rs

@@ -651,6 +651,10 @@ pub enum ContentBlockStart {
         input: serde_json::Value, // usually empty or partial
         #[serde(skip_serializing_if = "Option::is_none")]
         caller: Option<ToolCaller>,
+        /// Google thought signature, when the first streaming chunk of this
+        /// tool call carried `extra_content.google.thought_signature`.
+        #[serde(skip_serializing_if = "Option::is_none")]
+        thought_signature: Option<String>,
     },
     #[serde(rename = "server_tool_use")]
     ServerToolUse {

crates/tui/src/plugins/marketplace/mod.rs

@@ -13,5 +13,3 @@ pub mod parsers;
 #[cfg(test)]
 mod tests;
 pub mod types;
-
-pub use parsers::MarketplaceDocument;

crates/tui/src/plugins/marketplace/tests.rs

@@ -6,7 +6,7 @@
 
 use serde_json::Value;
 
-use super::MarketplaceDocument;
+use super::parsers::MarketplaceDocument;
 use super::parsers::parse_catalog;
 use super::types::{
     CatalogTier, MarketplaceCatalogId, MarketplaceFormat, MarketplaceInstallPlan,

crates/tui/src/purge.rs

@@ -636,6 +636,7 @@ mod tests {
                 name: name.to_string(),
                 input,
                 caller: None,
+                thought_signature: None,
             }],
         }
     }
@@ -827,6 +828,7 @@ mod tests {
                 name: "purge_context".to_string(),
                 input: json!({"operations": operations}),
                 caller: None,
+                thought_signature: None,
             }],
             model: "mock-model".to_string(),
             stop_reason: None,

crates/tui/src/rlm/session.rs

@@ -357,6 +357,7 @@ fn compact_content_block(block: &ContentBlock) -> Value {
             name,
             input,
             caller,
+            ..
         } => json!({
             "type": "tool_use",
             "id": id,

crates/tui/src/runtime_api/sessions.rs

@@ -622,6 +622,7 @@ pub(super) fn messages_from_thread_detail(detail: &ThreadDetail) -> Vec<Message>
                             name: tool_name,
                             input,
                             caller: None,
+                            thought_signature: None,
                         });
                     }
                 }
@@ -825,8 +826,7 @@ pub(super) fn session_to_detail(session: SavedSession) -> SessionDetailResponse
                         id,
                         name,
                         input,
-                        caller,
-                    } => {
+                        caller, ..} => {
                         let mut obj =
                             json!({ "type": "tool_use", "id": id, "name": name, "input": input });
                         if let Some(caller) = caller {

crates/tui/src/runtime_api/tests.rs

@@ -226,6 +226,7 @@ fn session_detail_tool_use_preserves_caller_metadata() {
                 caller_type: "subagent".to_string(),
                 tool_id: Some("parent-tool".to_string()),
             }),
+            thought_signature: None,
         },
     ]));
 

crates/tui/src/runtime_threads.rs

@@ -6103,6 +6103,7 @@ impl RuntimeThreadManager {
                                 name: tool_name,
                                 input,
                                 caller: None,
+                                thought_signature: None,
                             });
                         }
                     }

crates/tui/src/runtime_threads/tests.rs

@@ -3235,12 +3235,14 @@ async fn seed_thread_keeps_tool_results_on_preceding_turn() -> Result<()> {
                     name: "shell".to_string(),
                     input: json!({ "cmd": "one" }),
                     caller: None,
+                    thought_signature: None,
                 },
                 ContentBlock::ToolUse {
                     id: "tool-2".to_string(),
                     name: "shell".to_string(),
                     input: json!({ "cmd": "two" }),
                     caller: None,
+                    thought_signature: None,
                 },
             ],
         },

crates/tui/src/session_manager.rs

@@ -2545,6 +2545,7 @@ mod tests {
                 name: "read_file".to_string(),
                 input: serde_json::json!({"path": "README.md"}),
                 caller: None,
+                thought_signature: None,
             }],
         }];
         let session = create_saved_session(&messages, "test-model", tmp.path(), 0, None);
@@ -2579,6 +2580,7 @@ mod tests {
                 name: "read_file".to_string(),
                 input: serde_json::json!({"path": "README.md"}),
                 caller: None,
+                thought_signature: None,
             }],
         }];
         let session = create_saved_session(&messages, "test-model", tmp.path(), 0, None);
@@ -2615,6 +2617,7 @@ mod tests {
                 name: "read_file".to_string(),
                 input: serde_json::json!({"path": "README.md"}),
                 caller: None,
+                thought_signature: None,
             }],
         }];
         let session = create_saved_session(&messages, "test-model", tmp.path(), 0, None);
@@ -2678,6 +2681,7 @@ mod tests {
                         ]
                     }),
                     caller: None,
+                    thought_signature: None,
                 }],
             },
             Message {
@@ -2725,6 +2729,7 @@ mod tests {
                     name: "exec_shell".to_string(),
                     input: serde_json::json!({"command": "cargo test -p codewhale-tui"}),
                     caller: None,
+                    thought_signature: None,
                 }],
             },
             Message {
@@ -2777,6 +2782,7 @@ mod tests {
                     name: "exec_shell".to_string(),
                     input: serde_json::json!({"command": "cargo check"}),
                     caller: None,
+                    thought_signature: None,
                 }],
             },
             Message {

crates/tui/src/session_peek.rs

@@ -390,6 +390,7 @@ mod tests {
                     name: "read_file".to_string(),
                     input: serde_json::json!({ "path": "/etc/shadow" }),
                     caller: None,
+                    thought_signature: None,
                 },
                 ContentBlock::ToolResult {
                     tool_use_id: "call-1".to_string(),

crates/tui/src/tool_history_repair.rs

@@ -222,6 +222,7 @@ mod tests {
                 name: "read_file".to_string(),
                 input: json!({"path": "README.md"}),
                 caller: None,
+                thought_signature: None,
             }],
         }
     }

crates/tui/src/tool_output_receipts.rs

@@ -317,6 +317,7 @@ mod tests {
                 name: name.to_string(),
                 input,
                 caller: None,
+                thought_signature: None,
             }],
         }
     }

crates/tui/src/tools/subagent/advisor.rs

@@ -450,6 +450,7 @@ mod tests {
                     name: "exec_shell".to_string(),
                     input: serde_json::json!({"command": format!("echo {i}")}),
                     caller: None,
+                    thought_signature: None,
                 }],
             });
             // user message with ToolResult

crates/tui/src/tools/subagent/tests.rs

@@ -6538,6 +6538,7 @@ async fn small_surface_fork_context_survives_fresh_child_discovery() {
                     name: TOOL_SEARCH_NAME.to_string(),
                     input: json!({"query": "web"}),
                     caller: None,
+                    thought_signature: None,
                 }],
             },
             Message {

crates/tui/src/tui/history/tests.rs

@@ -1055,6 +1055,7 @@ fn history_replays_update_plan_tool_use_as_plan_card() {
                 ]
             }),
             caller: None,
+            thought_signature: None,
         }],
     };
 

crates/tui/src/tui/ui/dispatch.rs

@@ -90,6 +90,7 @@ pub(crate) fn push_assistant_message(
             name,
             input,
             caller: None,
+            thought_signature: None,
         });
     }
 

crates/tui/src/tui/ui/frame.rs

@@ -586,7 +586,7 @@ pub(crate) fn live_tool_receipt_messages(
     let mut messages = Vec::with_capacity(2);
     if let Some(tool_use_msg) = app.api_messages.iter().rev().find(|message| {
         message.content.iter().any(|block| {
-            matches!(block, ContentBlock::ToolUse { id: tool_use_id, .. } if tool_use_id == id)
+            matches!(block, ContentBlock::ToolUse { id: tool_use_id, ..} if tool_use_id == id)
         })
     }) {
         messages.push(tool_use_msg.clone());

crates/tui/src/tui/ui/session_state.rs

@@ -873,6 +873,7 @@ pub(crate) fn mirror_saved_api_key_in_config(
         ApiProvider::Meta => &mut providers.meta,
         ApiProvider::Xai => &mut providers.xai,
         ApiProvider::Mistral => &mut providers.mistral,
+        ApiProvider::Google => &mut providers.google,
         ApiProvider::Telecomjs => &mut providers.telecomjs,
         ApiProvider::ModelstudioTokenPlan => &mut providers.modelstudio_token_plan,
         ApiProvider::ModelstudioTokenPlanAnthropic => {

crates/tui/src/tui/ui/tests.rs

@@ -5282,6 +5282,7 @@ async fn tool_result_api_content_never_advertises_unowned_live_output_as_retriev
             name: "exec_shell".to_string(),
             input: serde_json::json!({"command": "cargo test"}),
             caller: None,
+            thought_signature: None,
         }],
     });
 
@@ -5314,6 +5315,7 @@ fn live_tool_receipt_messages_clones_only_matching_tool_use() {
             name: "exec_shell".to_string(),
             input: serde_json::json!({"command": "old"}),
             caller: None,
+            thought_signature: None,
         }],
     });
     app.api_messages.push(Message {
@@ -5332,6 +5334,7 @@ fn live_tool_receipt_messages_clones_only_matching_tool_use() {
             name: "read_file".to_string(),
             input: serde_json::json!({"path": "src/main.rs"}),
             caller: None,
+            thought_signature: None,
         }],
     });
 
@@ -5340,7 +5343,7 @@ fn live_tool_receipt_messages_clones_only_matching_tool_use() {
     assert_eq!(messages.len(), 2);
     assert!(matches!(
         &messages[0].content[0],
-        ContentBlock::ToolUse { id, name, .. } if id == "call-new" && name == "read_file"
+        ContentBlock::ToolUse { id, name, ..} if id == "call-new" && name == "read_file"
     ));
     assert!(matches!(
         &messages[1].content[0],
@@ -21471,6 +21474,7 @@ fn backtrack_cut_index_skips_tool_result_user_messages() {
                 name: "read_file".into(),
                 input: serde_json::json!({"path":"x"}),
                 caller: None,
+                thought_signature: None,
             }],
         },
         Message {

crates/tui/tests/integration/integration_mock_llm.rs

@@ -93,6 +93,7 @@ fn assistant_tool_call(id: &str, name: &str, input: serde_json::Value) -> Messag
             name: name.to_string(),
             input,
             caller: None,
+            thought_signature: None,
         }],
     }
 }

crates/tui/tests/integration/reasoning_content_replayed_after_tool_call.rs

@@ -33,6 +33,7 @@ fn assistant_thinking_tool_call(
                 name: name.to_string(),
                 input,
                 caller: None,
+                thought_signature: None,
             },
         ],
     }

docs/PROVIDERS.md

@@ -131,6 +131,7 @@ the listed provider env vars.
 | `telecomjs` | `[providers.telecomjs]` | OpenAI Chat Completions | `TELECOMJS_API_KEY` |
 | `xai` | `[providers.xai]` | OpenAI Chat Completions | `XAI_API_KEY` |
 | `mistral` | `[providers.mistral]` | OpenAI Chat Completions | `MISTRAL_API_KEY` |
+| `google` | `[providers.google]` | OpenAI Chat Completions (official Gemini OpenAI-compat route; captures and replays thought signatures on tool calls) | `GOOGLE_API_KEY`, `GEMINI_API_KEY` |
 | `modelstudio-token-plan` | `[providers.modelstudio_token_plan]` | OpenAI Chat Completions | `MODELSTUDIO_API_KEY`, `DASHSCOPE_API_KEY` |
 | `modelstudio-token-plan-anthropic` | `[providers.modelstudio_token_plan_anthropic]` | Anthropic Messages | `MODELSTUDIO_API_KEY`, `DASHSCOPE_API_KEY` |
 | `modelstudio-coding-plan` | `[providers.modelstudio_coding_plan]` | OpenAI Chat Completions | `MODELSTUDIO_API_KEY`, `DASHSCOPE_API_KEY` |
@@ -453,6 +454,7 @@ configuration path instead of guessing a vendor page.
 | `telecomjs` | [TelecomJS TokenHub](https://aigw.telecomjs.com/) |
 | `xai` | [xAI Console](https://console.x.ai/) for an API key, Codewhale-owned device login, or explicitly consented read-only Grok CLI credentials. |
 | `mistral` | [Mistral Console (la Plateforme)](https://console.mistral.ai/api-keys) |
+| `google` | [Google AI Studio](https://aistudio.google.com/apikey) — Codewhale uses the official Gemini OpenAI-compatible endpoint and never reads Google OAuth files. |
 | `modelstudio-token-plan`, `modelstudio-token-plan-anthropic`, `modelstudio-coding-plan`, `modelstudio-coding-plan-anthropic` | [Alibaba Cloud Model Studio (Bailian console)](https://bailian.console.aliyun.com/) — create or copy a Model Studio API key. |
 | `custom` | Set the named provider's `base_url` and `api_key_env` or `api_key`; no canonical vendor credential page exists. |
 
@@ -521,6 +523,7 @@ Kimi remains API-key-only; external consent for Kimi is rejected.
 | `siliconflow-CN` | `[providers.siliconflow_cn]` | `SILICONFLOW_API_KEY` | `SILICONFLOW_BASE_URL`; default `https://api.siliconflow.cn/v1` | Uses the SiliconFlow model set | China regional SiliconFlow route. Falls back to `[providers.siliconflow]` for api_key / base_url / model when unset. Select it with `provider = "siliconflow-CN"` or `CODEWHALE_PROVIDER=siliconflow-CN`. |
 | `arcee` | `[providers.arcee]` | `ARCEE_API_KEY` | `ARCEE_BASE_URL`; default `https://api.arcee.ai/api/v1` | `trinity-large-thinking`, `trinity-large-preview` | Arcee AI direct OpenAI-compatible route, tracked as 256K-context BF16 serving. `ARCEE_MODEL` is accepted. OpenRouter's `arcee-ai/trinity-large-thinking` remains the OpenRouter namespaced model ID; direct Arcee uses the bare `trinity-large-thinking` ID. |
 | `moonshot` | `[providers.moonshot]` | `MOONSHOT_API_KEY`, `KIMI_API_KEY` | `MOONSHOT_BASE_URL`, `KIMI_BASE_URL`; default `https://api.moonshot.ai/v1` | Direct Moonshot: `kimi-k3`, `kimi-k2.7-code`, `kimi-k2.6`; Kimi Code membership: `k3`, `kimi-for-coding`, `kimi-for-coding-highspeed` at `https://api.kimi.com/coding/v1` | Moonshot/Kimi route. `kimi` and `kimi-k2` aliases select `kimi-k2.7-code`; `MOONSHOT_MODEL`, `KIMI_MODEL_NAME`, and `KIMI_MODEL` are accepted. Kimi thinking streams through `reasoning_content`; Codewhale keeps it in Thinking cells and replays it for thinking/tool-call continuity. For direct K3, use exact `base_url = "https://api.moonshot.ai/v1"` and `model = "kimi-k3"`; it is always-thinking and receives top-level `reasoning_effort = "low" | "high" | "max"` (`off` normalizes to `low`), uses only `max_completion_tokens`, and omits `temperature`/`top_p` per the [K3 quickstart](https://platform.kimi.ai/docs/guide/kimi-k3-quickstart). For Kimi Code K3, use a key from the [Kimi Code console](https://www.kimi.com/code/console), exact `base_url = "https://api.kimi.com/coding/v1"`, and bare `model = "k3"`; `off` becomes enabled `low`, while normal dispatched `auto` selects and sends a concrete Codewhale tier. Only an omitted reasoning setting leaves the provider default in control. That membership route defaults safely to 262,144 context tokens; the [Kimi Code model-tier table](https://www.kimi.com/code/docs/en/kimi-code/models.html) grants Allegretto and higher plans up to 1M, which those plans may express as `context_window = 1048576`. `k3[1m]` is Claude Code-only and Codewhale rejects it. `kimi-for-coding` remains the valid K2.7 membership route, and `kimi-for-coding-highspeed` is its own high-speed roster entry (262,144 context); membership ids are rejected on the direct platform endpoint, and `kimi-k3` stays rejected on the membership endpoint. Billing is decided by the endpoint the route resolves to, judged once against the two exact product endpoints: direct Moonshot (`https://api.moonshot.ai/v1` or the default) bills metered with dollar estimates, the exact Kimi Code membership endpoint bills as Kimi Code quota and never shows dollar estimates, and anything else — a gateway host, a neighboring Kimi-hosted path — reports `cost: unknown` rather than borrowing either product. An imported Kimi Code token with no `base_url` in its table still resolves to the membership endpoint, so it bills as Kimi Code quota and never accrues dollars. A completed turn, parent or sub-agent, is billed from the immutable endpoint receipt its own client was built with, never from a later config re-read: `MOONSHOT_BASE_URL`/`KIMI_BASE_URL` are merged into the *active* provider's table only, and an in-turn provider switch can move the ambient config off the route that actually ran. Legacy `auth_mode = "kimi_oauth"` fails to API-key guidance without probing Kimi CLI files. Codewhale does not impersonate `kimi_cli` or `kimi_code_cli`. **China-region keys:** contributor field evidence (@vFONGv, PR #5229, verified on Windows 10) reports that a China-region Moonshot key must be paired with `base_url = "https://api.moonshot.cn/v1"`; left on the default international host (`https://api.moonshot.ai/v1`) it fails authentication. We have no China-region key to verify this ourselves, so it is recorded as a user report rather than a tested route. Note also that editing `base_url` alone does not take effect until `codewhale auth set` is re-run for that provider. |
+| `google` | `[providers.google]` | `GOOGLE_API_KEY`, `GEMINI_API_KEY` | `GOOGLE_BASE_URL`, `GEMINI_BASE_URL`; default `https://generativelanguage.googleapis.com/v1beta/openai/` | `gemini-3.1-pro-preview` (default); `/model` also lists `gemini-3-pro-preview`, `gemini-3.6-flash`, `gemini-3.5-flash`, `gemini-3.5-flash-lite`, `gemini-2.5-pro`, `gemini-2.5-flash` | Google Gemini as its own backend on the official OpenAI-compatible Chat Completions route. Thinking models capture `extra_content.google.thought_signature` on tool calls and replay it with the assistant tool-call messages; replaying a tool call whose signature was not captured fails closed with an actionable error instead of letting the tool loop break. `gemini-2.5-flash-lite` ships thinking off and degrades with a warning instead. Reasoning effort maps onto the documented `google.thinking_config.thinking_level` (`low`/`high`). The dialect binds to the exact official base URL: a `google` row pointed at another gateway gets plain OpenAI semantics and no signature requirements. Codewhale never reads Google OAuth files; only an AI Studio API key is used. Not live-tested against the real endpoint in this environment. |
 | `zai` | `[providers.zai]` | `ZAI_API_KEY`, `Z_AI_API_KEY` | `ZAI_BASE_URL`, `Z_AI_BASE_URL`; default `https://api.z.ai/api/coding/paas/v4`; general API `https://api.z.ai/api/paas/v4` | `GLM-5.2` default; `/model` also lists `GLM-5.3`, `GLM-5.1`, and `GLM-5-Turbo` | Z.AI GLM Coding Plan route. `GLM-5.2` stays the default. `GLM-5.3` is a first-class picker row (`model = "GLM-5.3"` or `ZAI_MODEL=GLM-5.3`) and is not rewritten to another vendor. Limits and reasoning options are inherited from `GLM-5.2` until Z.ai publishes distinct 5.3 metadata; it carries no price. A live call can still 429 with entitlement code 1311 on accounts that are not provisioned for 5.3. |
 | `stepfun` | `[providers.stepfun]` | `STEPFUN_API_KEY`, `STEP_API_KEY` | `STEPFUN_BASE_URL`, `STEP_BASE_URL`; default `https://api.stepfun.ai/v1`; Coding Plan endpoint `https://api.stepfun.ai/step_plan/v1` | `step-3.7-flash` | StepFun / StepFlash direct OpenAI-compatible route. `/provider` setup asks which billing route the key belongs to — pay-as-you-go or Step Plan — validates the key against the chosen endpoint, and writes the answer to `[providers.stepfun].base_url` only. A base URL that is neither recognized route is left alone and the question is skipped. You can also set `[providers.stepfun].base_url` or `STEP_BASE_URL` to the Coding Plan URL by hand. Offline accounting labels recognized routes as `stepfun-payg` or `stepfun-plan` without persisting the raw endpoint, and only the standard PAYG route receives token pricing. `STEPFUN_MODEL` and `STEP_MODEL` are accepted. |
 | `minimax` | `[providers.minimax]` | `MINIMAX_API_KEY` | `MINIMAX_BASE_URL`; default `https://api.minimax.io/v1`; China `https://api.minimaxi.com/v1` | `MiniMax-M3`, `MiniMax-M2.7`, `MiniMax-M2.7-highspeed`, `MiniMax-M2.5`, `MiniMax-M2.5-highspeed`, `MiniMax-M2.1`, `MiniMax-M2.1-highspeed`, `MiniMax-M2` | MiniMax direct OpenAI-compatible route. Codewhale sends `reasoning_split = true` so MiniMax thinking arrives separately from answer text. Both MiniMax dialects sell pay-as-you-go and Token Plan over the same endpoints and the same key, so billing is classified from the credential *product*, never from the endpoint or from a default. `mode = "token-plan"` in `[providers.minimax]`/`[providers.minimax_anthropic]`, or a Token Plan key shaped `sk-cp…`, bills as MiniMax Token Plan quota with no dollar estimates; an explicit pay-as-you-go mode (`pay-as-you-go`/`payg`/`metered`) wins over key shape. The key's product prefix is only visible when the key is in config, bound by `api_key_env`, or exported as `MINIMAX_API_KEY` on an official endpoint — a key saved through `codewhale auth set` (secret store / OS keyring) is deliberately not read to classify billing. With no explicit mode and no visible product marker the route reports `cost: unknown` rather than assuming pay-as-you-go, so a Token Plan account is never charged invented dollars. Custom/gateway endpoints also fail closed with `cost: unknown`. Official M3 input modalities are text, image, and video; M2.7 is text-only. |