Ship AI features fast — with a clear record of what your chains and agents actually did. One .wrap() call, low overhead, and answers you can show investors or customers.
No credit card required • Free tier to get started
When a chain misfires at 2am, you need the input, tool steps, and final output — not a wall of unstructured logs.
Founders, customers, and early enterprise pilots want proof your AI did the right thing — without a compliance team on day one.
Start with hashes-only tracing today. Turn on full context when a deal or regulator asks — same SDK, no rewrite.
pip install "audifact-sdk[langchain]" "langchain-core>=0.3.0"
python sdk/examples/enterprise/langchain_basic_and_agent.py
AUDIFACT_API_KEY=af_YOUR_KEY python ...
Also works with LlamaIndex, LangGraph, CrewAI, and AutoGen — same explorer, same event model.
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.output_parsers import StrOutputParser
from audifact.langchain import AudifactLangChainTracer
chain = prompt | llm | StrOutputParser()
tracer = AudifactLangChainTracer(
api_key="af_YOUR_API_KEY",
agent_name="product-copilot",
)
traced = tracer.wrap(chain)
result = traced.invoke("Summarize our refund policy for the user.")
# → chain_execution + chain_responseWorks with any Runnable — RetrievalQA-style flows, structured Pydantic outputs, and more
from langchain.agents import AgentExecutor, create_tool_calling_agent
from audifact.langchain import AudifactLangChainTracer
agent = AgentExecutor(agent=..., tools=[search, calc], verbose=True)
tracer = AudifactLangChainTracer(
api_key="af_YOUR_API_KEY",
agent_name="support-agent",
include_context=True, # tool_calls in output_json
)
traced = tracer.wrap(agent)
traced.invoke({"input": "Look up plan limits and draft a reply."})
# → agent_execution + agent_responseAlso supports LlamaIndex agents, CrewAI crews, LangGraph apps — same event pairs
Safe demo — run without an API key locally. Add AUDIFACT_API_KEY when you're ready to persist events.
Every run becomes a linked decision + outcome. Readable public_decision fields in the explorer — no parsing raw JSON to understand what happened.
One tracer, one .wrap() — no custom logging infra. Works with your existing LangChain stack.
Free tier to start. Default hashes-only mode keeps payloads small; opt in to rich context when you need it.
Don't ship customer data to a third-party log sink. Capture proofs, not prompts — unless you choose otherwise.
Same SDK from prototype → Series A pilot → enterprise deal. LlamaIndex, LangGraph, CrewAI ready when you are.
Free API key. Five minutes to your first signed event. No enterprise sales call required.
Questions? Check the quickstart or enterprise pages when you scale up.