instrument: orchestration framework adapters (M1.C part 1)#96
Closed
mmercuri wants to merge 2 commits into
Closed
Conversation
Ports the six orchestration-tier framework adapters from the ateam
reference implementation onto the new layerlens.instrument base layer:
LangChain, LangGraph, CrewAI, AutoGen, Langfuse, Agentforce
These six frameworks share heavy multi-file packages (lifecycle,
handoff, state, tools, llm, nodes) and benefit from being landed
together for cross-cutting tests. The lighter agent-tier adapters
(Agno, OpenAI Agents, Pydantic-AI, etc.) ship in a follow-up PR
(M1.C part 2).
Scope
-----
- src/layerlens/instrument/adapters/frameworks/{langchain,langgraph,
crewai,autogen,langfuse,agentforce}/: per-framework packages
- tests/instrument/adapters/frameworks/{test_autogen,test_crewai,
test_langfuse}_adapter.py: unit tests where the framework is
importable on the test runner; langchain / langgraph / agentforce
ride the bulk-ported smoke harness in part 2
- samples/instrument/{langchain,langgraph,crewai,autogen,
agentforce}/: runnable samples
- docs/adapters/frameworks-{langchain,langgraph,crewai,autogen,
agentforce}.md: per-framework integration guide
- pyproject.toml: six new optional extras
(langchain, langgraph, crewai, autogen, agentforce, langfuse-importer)
with python_version markers where the upstream framework requires
3.9+ or 3.10+; pyright/ruff exclusions for the dynamic monkey-
patching framework code
Blast radius
------------
- Default `pip install layerlens` install set is unchanged. Each
framework's heavy deps (langchain, crewai, etc.) are gated behind
their own extra.
- No changes to existing public API surface.
- Importing layerlens.instrument still does NOT pull in any framework
module (lazy registry lookup).
Test plan
---------
- uv run pytest tests/instrument/adapters/frameworks/ -x ->
37 passed (autogen + crewai + langfuse units)
- langchain / langgraph / agentforce coverage continues in part 2 via
the bulk-ported smoke harness
Stacks on
---------
- feat/instrument-base-foundation (M1.A) — required for the
BaseAdapter surface this PR consumes.
LAY-3400 umbrella (M1.C part 1).
Contributor
Author
|
Linear: https://linear.app/layerlens/issue/LAY-3400 (Framework adapters part 1 — orchestration tier: LangChain, LangGraph, CrewAI, AutoGen, Langfuse, Agentforce). Stacked on PR #93. Under Apollo M1 epic LAY-3423. |
This was referenced Apr 26, 2026
deferred) LayerLensCallbackHandler implements serialize_for_replay (returns a non-stub ReplayableTrace populated with self._trace_events plus the capture config) AND a working execute_replay coroutine, but get_adapter_info().capabilities did not declare AdapterCapability.REPLAY. The atlas-app catalog UI reads that list to surface replay support, so customers were told they could not replay traces from LangChain even though the adapter supports it. PR #119 (brand leak + capability declarations) wired REPLAY for the adapters that lived on its branch; LangChain was deferred because it lives on the orchestration source-port branch (PR #96). This closes that deferral per CLAUDE.md item 5/11. STREAMING is intentionally NOT declared. Per CLAUDE.md 'no fake claims', a capability is only declared if the adapter actually implements it. The LangChain adapter registers on_chat_model_start / on_llm_start / on_llm_end / on_tool_* / on_agent_* / on_chain_* — but NOT on_llm_new_token, the LangChain streaming callback. No per-chunk events flow through this adapter. Adding STREAMING here would mislead the catalog UI into telling customers they can stream from an adapter that does not see chunks. Tests: added tests/instrument/adapters/frameworks/test_langchain_capabilities.py with three regressions: * test_declares_replay_capability — REPLAY surfaces via info().capabilities * test_does_not_declare_streaming_capability — STREAMING stays absent until on_llm_new_token is wired and tested explicitly * test_get_adapter_info_matches_info_wrapper — info() and get_adapter_info() agree on the capability list Verification: * uv run --with pytest python -m pytest tests/instrument/adapters/frameworks/test_langchain_capabilities.py -x -> 3 passed * uv run --with pytest python -m pytest tests/instrument/adapters/frameworks/ -x -> 40 passed (no regressions in autogen / crewai / langfuse suites)
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Ports the six orchestration-tier framework adapters from the ateam
reference implementation onto the new
layerlens.instrumentbaselayer:
LangChain, LangGraph, CrewAI, AutoGen, Langfuse, Agentforce
These six frameworks share heavy multi-file packages (lifecycle,
handoff, state, tools, llm, nodes) and benefit from being landed
together for cross-cutting tests. The lighter agent-tier adapters
(Agno, OpenAI Agents, Pydantic-AI, etc.) ship in a follow-up PR
(M1.C part 2).
Scope
src/layerlens/instrument/adapters/frameworks/{langchain,langgraph,crewai,autogen,langfuse,agentforce}/— per-framework packages
tests/instrument/adapters/frameworks/{test_autogen,test_crewai,test_langfuse}_adapter.py— unit tests where the framework is importable on the test runner;
langchain/langgraph/agentforceride the bulk-ported smokeharness in part 2
samples/instrument/{langchain,langgraph,crewai,autogen,agentforce}/— runnable samples
docs/adapters/frameworks-{langchain,langgraph,crewai,autogen,agentforce}.md— per-framework integration guide
pyproject.toml— six new optional extras (langchain,langgraph,crewai,autogen,agentforce,langfuse-importer) withpython_versionmarkers where the upstream framework requires3.9+ or 3.10+; pyright/ruff exclusions for the dynamic monkey-
patching framework code
Blast radius
pip install layerlensinstall set is unchanged. Eachframework's heavy deps (langchain, crewai, etc.) are gated behind
their own extra.
layerlens.instrumentstill does NOT pull in anyframework module (lazy registry lookup).
Test plan
uv run pytest tests/instrument/adapters/frameworks/ -x—37 passed (autogen + crewai + langfuse units)
langchain/langgraph/agentforcecoverage continues inpart 2 via the bulk-ported smoke harness
m-peko) verifies callback/lifecycle wiringStacks on
feat/instrument-base-foundation(M1.A) — required for theBaseAdaptersurface this PR consumes.Linear
LAY-3400 umbrella (M1.C part 1).