Kitaru LangSmith importer

Import LangSmith run-query and bulk-export records as Kitaru sessions. This package backs the built-in kitaru/langsmith importer and runs on a Kitaru worker, so the export is parsed in your environment.

Most users do not install or call this package directly. Start a Kitaru worker, then select the built-in importer:

kitaru session import langsmith-runs.jsonl \
  --importer kitaru/langsmith@latest \
  --agent support-agent@latest \
  --wait

The importer accepts LangSmith JSON and JSONL export shapes, reconstructs run hierarchy, and groups traces using thread-like metadata or an explicit grouping path. Re-importing the same source identity skips sessions that Kitaru already stores.

See the LangSmith import guide for accepted formats, grouping parameters, deduplication behavior, and fidelity limits.

Validation limits

Nested node trees support at most 64 nodes along a parent path, counting the root as level 1. Tool-activity scans have a separate limit of 64 container or embedded-JSON decoding steps. Costs must be finite and nonnegative, token counts must be nonnegative, and returned sessions must serialize as JSON.

Duplicate run IDs reject the complete LangSmith trace before grouping, including identical repeated records. Reusing a run ID in another trace remains valid; a self-parent link is treated as a root. A failure discovered before grouping leaves other traces available, even when they share a session key. Invalid fields or serialization failures found after grouping reject that grouped session; unrelated sessions still import.

Adapter

Install the package with the adapter extra to use the adapter, which adds the provider SDK it needs:

uv add "kitaru-langsmith-importer[adapter]"

The package also ships an adapter that imports LangSmith traces of wrapped agent runs. The adapter uses the LangSmith SDK already configured in your process and the Kitaru connection from your environment. Set LANGSMITH_API_KEY (plus LANGSMITH_ENDPOINT for a self-hosted instance) to the credentials the trace fetch authenticates with, then wrap your agent entrypoint in a LangSmithAdapter and run it through the adapter.

from kitaru_langsmith_importer.adapter import LangSmithAdapter

adapter = LangSmithAdapter()
result = adapter.run(my_agent, "Hello")

The adapter runs the function inside a LangSmith trace, waits for LangSmith to finish ingesting the trace, fetches it, and imports it as one Kitaru session. Use run_async for async functions. When the trace does not complete within the completeness timeout, the adapter creates a failed session carrying the trace id.

Links

• Kitaru documentation
• Source code
• Issue tracker

Licensed under Apache-2.0.