HOW IT WORKS

Wrap your agent. Get the full trace tree. Fix it in minutes.

No manual instrumentation, no separate spans to write. AgentLogs hooks into your client or callback once — every LLM call, tool call, and step after that is captured and replayable.

1

Wrap your client

One import. No manual span creation, no decorators on every function.

2

Run your agent as usual

Every LLM call, tool call, and chain step is captured in the background.

3

See the full trace tree

Nested spans show exactly which step failed, how long each took, and what it cost.

4

Replay and fix

Re-run any step in isolation with its original input to confirm a fix before shipping.

$ pip install agentlogs
from agentlogs import wrap_openai
from openai import OpenAI

client = wrap_openai(OpenAI())

# every call below is traced automatically —
# no extra lines at the call site
resp = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "..."}],
)

Don't see your framework? Check integrations — the raw HTTP API works with anything.

trace tree · live preview
[✓] agent:planner — tool_call: search_web120ms
[✓] agent:executor — llm_call: gpt-4o340ms
[✕] agent:validator — tool_call: check_output80ms

Simulated for this demo — real traces from your project appear the same way in Traces.

What tracing gives you

  • Full nested trace tree — every LLM call, tool call, and chain step in one view, not scattered logs
  • Per-step cost and token counts, so you know which call in the pipeline is expensive
  • Replay any step in isolation with its original input, to confirm a fix before re-running the whole agent
  • Auto-instrumentation for OpenAI + LangChain today — zero manual span code
Start tracing your agent →