Skip to content

SUBMIT.md

Latest commit

 

History

History
83 lines (71 loc) · 4.05 KB

SUBMIT.md

File metadata and controls

83 lines (71 loc) · 4.05 KB
Raw

langchain-anakin — verification status

Live-verified, not assumed

  • anakin-sdk is live on PyPI right now (confirmed via curl https://pypi.org/pypi/anakin-sdk/json → 200, version 0.1.0, published 2026-04-28, repository field correctly points to Anakin-Inc/anakin-py). pyproject.toml depends on it normally (anakin-sdk>=0.1.0,<0.2.0) — no local path override needed or present.

  • A standalone package is the upstream-endorsed path, not a workaround. Anakin-Inc/langchain-community is a real fork of langchain-ai/langchain-community and already has three closed PRs against the real upstream monorepo: #552, #553, #554 (all opened 2026-02-23, #554 stayed open until 2026-04-11). A LangChain maintainer closed #554 with:

    We no longer accept additional integrations in the langchain monorepo. [...] I suggest to: Create your own repository to distribute LangChain integrations, Publish the package to PyPI.

    So this package's approach — standalone repo, publish to PyPI as langchain-anakin — is exactly what upstream told the team to do, not a parallel/duplicate effort. The feat/anakin-integration branch still exists on the fork with the original monorepo-shaped code (libs/community/langchain_community/{document_loaders,tools,utilities}/anakin*) if anyone wants to reference the original implementation, but it should not be resubmitted upstream in that form.

  • All source files (langchain_anakin/*.py, tests/test_tools.py) pass python3 -m py_compile — syntactically valid.

  • Every field/method referenced against anakin-py's real source, not assumed: Anakin.__init__(api_key=, *, base_url=, ...), client.scrape(), client.search(), client.agentic_search() signatures, and the Document/SearchResult/SearchResultItem/AgenticSearchResult/ AgenticSearchData Pydantic model fields all read directly from anakin-py/src/anakin/client.py and models.py — not the public docs (which, per the Pipedream/n8n work in this same session, were found to omit real parameters like forceFresh and agentic search's useBrowser/ schema). Both are included here.

  • Tool structure follows langchain-exa's real, current, in-tree pattern (pulled from langchain-ai/langchain's actual repo via gh api, not memory) — BaseTool subclass, model_validator-based client construction, _run with typed args mirroring the underlying client call.

What's NOT verified

pip install hung in this environment with near-zero CPU usage after ~10 minutes (checked via ps — the process was idle, not computing), consistent with restricted network access in this sandbox rather than a real problem with the dependencies. Never got a live pytest run. Before trusting this package:

python3 -m venv .venv && source .venv/bin/activate
pip install -e ".[test]"
pytest

The five tests in tests/test_tools.py mock the Anakin client directly (no live API calls, no network needed beyond the pip install step) — they should pass immediately once the environment can actually install langchain-core and anakin-sdk (both are normal PyPI installs now).

Steps (needs the account owner)

  1. Run the install + test commands above in an environment with normal network access.
  2. Push as Anakin-Inc/anakin-langchain, publish to PyPI as langchain-anakin.
  3. Optional, once live: submit to LangChain's own docs (docs.langchain.com) as a community/partner integration listing — not investigated in this pass, same-shaped question as the Google ADK / Agno submissions elsewhere in this session (a docs PR, likely to langchain-ai/langchain or langchain-ai/docs).

Not done

Never installed or run — verified by static reading against real source only. Treat as reviewed-but-unexecuted, same status as the Pipedream components before this note was written for them.