Examples
Build the package once (npm run build from repo root) and then run any example with tsx. Each folder ships a README with extra notes.
| Folder | Capability | Notes |
|---|---|---|
basic/ |
Minimal loop | Single tool call with a real model (OpenAI). |
tools/ |
Multiple tools + events | Includes Tavily search integration and onEvent logging. |
tool-limit/ |
Global tool cap | Shows finalize system notice injection when limit is hit. |
structured-output/ |
Zod finalize | Demonstrates outputSchema with parsed JSON results. |
todo-planning/ |
Planning discipline | Enforces TODO workflow via useTodoList: true. |
summarization/ |
Token threshold | Triggers summarization based on limits.maxToken. |
summarize-context/ |
Summary + raw retrieval | Uses get_tool_response to fetch archived outputs. |
rewrite-summary/ |
Post-summary continuation | Continues conversation after context has been summarized. |
multi-agent/ |
Delegation | Wraps an agent as a tool to answer sub-questions. |
handoff/ |
Runtime handoff | Transfers control between specialist agents. |
mcp-tavily/ |
MCP tools | Discovers and uses MCP-hosted Tavily search. |
vision/ |
Multimodal | Sends text + image message parts through the adapter. |
Required environment variables:
OPENAI_API_KEY– needed for all OpenAI-backed examples.TAVILY_API_KEY– optional; enables real Tavily results intools/and the MCP example.
Tip: enable tracing.enabled: true while testing and review logs/<session>/trace.session.json for full event history, payloads, and usage details.