Model names require provider prefix for proper resolution.
Changed from x-ai/grok-4.1-fast to openrouter/x-ai/grok-4.1-fast.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Two issues fixed:
1. SSE event format mismatch (app.ts): Server was sending event type in
SSE event: field and only properties in data: field. SDK only yields
data: field content, so run command couldn't determine event types.
Fix: Include both type and properties in data field.
2. Agent validation in attach mode (run.ts): When using --attach, code
tried to validate agents locally via Agent.get() which requires
Instance context. But attach mode doesn't call bootstrap() so no
context exists, causing "No context found for instance" error.
Fix: Skip local agent validation in attach mode (server validates),
only validate locally in bootstrap mode where Instance context exists.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Updates tiara to 09b8f65 which includes:
- Fix AgentCoreClient error handling for validation errors
- Fix QdrantStore API limit compliance (500 max)
- Add integration test for hive-mind + agent-core daemon
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Fix SQLite binding types in usage storage with proper any cast
- Fix assistantMessage.time.start → .created in processor
- Handle legacy string model format in prompt.ts
- Extend SDK Agent type with fallback property for TUI
- Extend SDK KeybindsConfig with model_fallback_toggle
- Fix text/span styling for bold in prompt component
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Update stanley to 93240c9 (observability, circuit breaker, forms)
- Update zee with telegram user mode fixes
- Update tiara with test reorganization
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Update AI SDK packages
- Update openai-compatible SDK tool response handlers
- Fix test fixtures for models API
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add Tiara namespace configuration types
- Add process registry types for agent-core centralization
- Update persona config handling
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add codexbar tool for usage and cost tracking queries
- Add splitwise tool for expense management
- Integrate new tools into zee domain
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add MCP passthrough routes for tool execution
- Improve session message handling with better logging
- Add usage route to server
- Update daemon CLI with usage tracking integration
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add usage tracking module with pricing, storage, and cost calculation
- Add PROVIDER_SUPPORTED_PARAMS to filter unsupported parameters per provider
- Fix parseModel to handle edge cases in model string parsing
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Claude Opus 4.5 thinking mode requires topP >= 0.95 or unset.
The previous top_p: 0.88 could cause issues with extended thinking.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add tests for persona thinking configurations:
- Zee (GLM-4.7): preserved thinking mode via zai/zhipuai provider
- Stanley (Grok 4.1): reasoningEffort variants via xAI
- Johny (Claude Opus 4.5): thinkingConfig via Google/Antigravity
- Fallback models: Gemini 3 thinkingLevel high
- Fix bunfig.toml to only run tests from test/ directory (excludes dist/)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Zee (GLM-4.7): Enable preserved thinking mode (clear_thinking: false)
- Stanley (Grok 4.1): Set reasoningEffort to high
- Johny (Claude Opus 4.5): Configure thinkingBudget to 64000 tokens
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Add support for Tiara hive-mind to use agent-core as the single source
of truth for all data via Qdrant vector database.
Changes:
- src/memory/types.ts: Add TiaraNamespaces constants and entry types
- TiaraSwarmEntry, TiaraAgentEntry, TiaraTaskEntry
- TiaraCommunicationEntry, TiaraConsensusEntry, TiaraMetricsEntry
- TiaraDecisionEntry, TiaraLearningPattern
- Helper functions for creating memory inputs
- packages/agent-core/src/process/: Add process registry module
- registry.ts: ProcessRegistry class for agent/swarm tracking
- types.ts: ProcessInfo, ProcessStats types
- Heartbeat monitoring and status management
- packages/agent-core/src/server/route/: Add API routes
- memory.ts: Memory CRUD and search endpoints
- process.ts: Process registry endpoints with SSE events
- Update vendor/tiara submodule to use QdrantStore
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Add health status indicators to the TUI footer:
- Internet connectivity (◉ NET) with green/red/muted states
- Connected LLM providers count (◈ N LLM)
Also update CLAUDE.md to document both installation methods
(bun global install and manual ~/bin/agent-core).
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Read version from packages/agent-core/package.json instead of
generating dynamic timestamp versions. This allows setting a
permanent version by editing package.json.
Priority order:
1. AGENT_CORE_VERSION env var
2. OPENCODE_VERSION env var
3. package.json version (new)
4. Dynamic timestamp version
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Use permissive types for provider factories (multiple @ai-sdk/provider versions)
- Add @ts-expect-error for LanguageModel/LanguageModelV3 mismatches
- Fix tool type comparisons for "provider-defined" at runtime
- Filter out tool-approval-request/response from content arrays
- Add proper inputTokenDetails/outputTokenDetails to test fixtures
- Cast Model types for SDK compatibility in share-next
- Fix PromiseLike.catch usage in prompt.ts
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Make toModelMessage async to handle AI SDK v6's async convertToModelMessages
- Update config tests to use .agent-core instead of .opencode
- Update tool registry tests to use .agent-core instead of .opencode
- Update all callers of toModelMessage to await the result
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Artur Do Lago