Chat context

Status: implemented.

Use llm.ChatContext to carry model-visible conversation state across turns.

The context can contain messages, tool/function items, config updates, and handoff records. Provider paths use it to build the request sent to an LLM or realtime model.

Read and update safely

Agent.ChatContext() returns a read-only copy so callers can inspect history without mutating the agent's internal state. To replace the agent-owned context, call Agent.UpdateChatContext or Agent.UpdateChatCtx.

When tools are configured, chat-context copy logic can filter function items against the active tools. That keeps stale or invalid function state from being carried into a new context unless you explicitly opt out.

When to use it

• seed an agent with prior conversation
• carry tool-call results into the next model turn
• inspect session history in tests
• create parity evidence for conversation behavior

Evidence:

• core/llm/chat_context.go
• core/llm/chat_context_test.go
• core/llm/remote_chat_context.go
• core/agent/agent_test.go