agentlens-kk 0.1.2

AgentLens Python SDK — fleet governance for multi-framework agent stacks

AgentLens Python SDK

Fleet governance for multi-framework agent stacks.

pip install agentlens-sdk                     # core
pip install "agentlens-sdk[langchain]"        # + LangChain
pip install "agentlens-sdk[crewai]"           # + CrewAI
pip install "agentlens-sdk[autogen]"          # + AutoGen
pip install "agentlens-sdk[all]"              # everything

API: https://agentlens-api-production.up.railway.app

---

Quick start (any framework)

import agentlens

agentlens.init("al_live_xxx")  # or set AGENTLENS_API_KEY env var

with agentlens.trace("my-agent") as run:
    result = my_agent.run(input)
    run.tool_call("search", input="query", output="results")
    run.complete(result)

Decorator style

import agentlens

agentlens.init("al_live_xxx")

@agentlens.monitor
def my_agent(input):
    return result

# or with options:
@agentlens.monitor(agent_id="my-agent")
async def my_async_agent(input):
    return result

Record tools without a trace context

agentlens.record_tool("send_email", input="to: [e-mail]", output="sent")

---

LangChain (auto-instrumented)

import agentlens

agentlens.init("al_live_xxx")
# LangChain detected and instrumented automatically
agent.invoke({"input": "..."})

Or manually:

from agentlens import AgentTracer
from agentlens.langchain import AgentLensCallbackHandler

tracer = AgentTracer(api_key="...", api_url="https://agentlens-api-production.up.railway.app", agent_id="my-agent")
handler = AgentLensCallbackHandler(tracer)
result = executor.invoke({"input": "..."}, config={"callbacks": [handler]})
tracer.shutdown()

CrewAI

from agentlens.crewai import AgentLensCrewListener

listener = AgentLensCrewListener(api_key="...", api_url="https://agentlens-api-production.up.railway.app", agent_id="my-crew")
crew = Crew(agents=[...], tasks=[...], event_listeners=[listener])
crew.kickoff()

AutoGen

import agentlens

agentlens.init("al_live_xxx")
agentlens.wrap_agent(assistant)   # registers hooks on the agent instance
chat_result = assistant.initiate_chat(user_proxy, message="...")

Or manually:

from agentlens.autogen import AgentLensAutoGenHook, wrap_agent

hook = AgentLensAutoGenHook(api_key="...", api_url="https://agentlens-api-production.up.railway.app", agent_id="my-agent")
wrap_agent(assistant, hook)
chat_result = assistant.initiate_chat(user_proxy, message="...")
hook.tracer.shutdown()

---

Event types

All event types match packages/core/src/schema.ts:

Category	Types
Agent	agent.started agent.completed agent.failed agent.paused
Step	step.started step.completed step.failed step.retried
Tool	tool.called tool.returned tool.errored
LLM	llm.request llm.response llm.hallucination_suspected
Memory	memory.read memory.write
HITL	hitl.checkpoint_created hitl.approved hitl.rejected hitl.timeout
Cost	cost.token_usage