Checked other resources

• This is a feature request, not a bug report or usage question.
• I added a clear and descriptive title that summarizes the feature request.
• I used the GitHub search to find a similar feature request and didn't find it.
• I checked the LangChain documentation and API reference to see if this feature already exists.
• This is not related to the langchain-community package.

Package (Required)

• langchain-core

Feature Description

Work Ledger is an open-source library (MIT) for recording, replaying, and comparing LLM agent runs. It already ships a WorkLedgerCallbackHandler that inherits from langchain_core.callbacks.BaseCallbackHandler and captures:

• LLM/chat model calls with token usage
• Tool invocations with inputs/outputs
• Retriever queries with documents
• Chain-level inputs/outputs
• Causal links between steps

from work_ledger import WorkLedger, WorkLedgerCallbackHandler
from langchain_openai import ChatOpenAI

ledger = WorkLedger(store="./runs")
handler = WorkLedgerCallbackHandler(ledger, run_name="my-chain")

chain.invoke({"question": "hi"}, config={"callbacks": [handler]})
run = handler.get_run()  # structured Run with steps, metrics, causal links

After recording runs, you can diff them:

from work_ledger.testing.diff import RunDiff

diff = RunDiff(run_v1, run_v2)
print(f"Similarity: {diff.similarity:.0%}")
print(f"Token delta: {diff.token_diff:+d}")
print(f"Steps added: {diff.steps_added}")

Use Case

When developing LangChain applications, prompt changes, model swaps, or tool modifications can silently alter behavior. Currently there's no standard way to:

1. Record a chain/agent execution as a structured artifact
2. Compare two runs to see exactly what changed (steps, outputs, tokens, cost)
3. Set up golden-file regression tests for CI

Work Ledger fills this gap. It's not an observability platform — it's a testing/debugging tool that complements LangSmith.

Typical workflow:

• Record a "known good" run
• Make changes (prompt, model, tools)
• Record the new run
• RunDiff shows exactly what changed
• CLI: work-ledger diff <run1> <run2>

Proposed Solution

I'd like to propose adding Work Ledger to LangChain's community integrations documentation, so users can discover it as a testing tool.

The integration is already working — the WorkLedgerCallbackHandler properly inherits from BaseCallbackHandler, passes isinstance checks, and works with LCEL chains via RunnableConfig.

Tested with real OpenAI API calls through langchain-openai:

• Simple LLM calls ✓
• LCEL chains (prompt | llm | parser) ✓
• Tool-calling LLMs ✓
• Cross-run diff ✓

Alternatives Considered

• LangSmith: Great for observability/monitoring but focuses on tracing, not structured regression testing with diffs
• Manual scripts: Every team builds their own — no standard approach
• agent-vcr / agentgraph: Similar tools but single-framework; Work Ledger supports LangChain, LangGraph, PydanticAI, CrewAI, LlamaIndex, OpenAI SDK, Anthropic SDK

Additional Context

• Repository: https://github.com/metawake/work-ledger
• License: MIT
• 276 tests passing
• Also integrates with LangGraph via wrap_graph()
• No heavy dependencies — langchain-core is optional (handler falls back to plain class)