AgentTrace
AgentTrace Beta is live

One timeline, every agent execution.

AgentTrace records every tool call, state change, and LLM decision. It intelligently builds a deterministic timeline you can replay and debug.

No credit card required.

trace_id: tr_9f82a1b...
fetch_user_tickets
142ms
{
  "user_id": "usr_9921",
  "status": "open"
}
generate_replyLLM
1.2s

"Based on the open tickets, I will draft a response explaining the current downtime and offering a credited extension..."

Integrations
Python
TypeScript
LangChain
LlamaIndex
OpenAI
Gemini
Anthropic
GitHub Actions
Python
TypeScript
LangChain
LlamaIndex
OpenAI
Gemini
Anthropic
GitHub Actions
Python
TypeScript
LangChain
LlamaIndex
OpenAI
Gemini
Anthropic
GitHub Actions

AgentTrace — Your Best Choice for
Debugging Complex Agents

See how we stack up against traditional logging and specialized tools.

Features
AgentTrace
🦜
LangSmith
Sentry
Replay.io
Deterministic Replay
Yes (Browser only)
Execution Timeline
Full state & side-effects
Basic trace tree
Error stack traces
Time travel debugger
State Injection / Branching
Target Workloads
Agentic AI / Python
LLM Apps
General Web/Backend
Frontend JS
Cost
Free & Open Source Core
Paid SaaS
Paid SaaS
Paid SaaS
Setup
pip install & 1 line of code
Requires API keys & wrappers
SDK Integration
Custom Browser setup
Time Travel

Replay any execution, with 100% precision.

Don't guess what went wrong. AgentTrace intercepts all tool calls, LLM requests, and external state changes, allowing you to replay failed agents locally exactly as they ran in production.

  • No mocking required for replays.
  • Cryptographically verified states.
00:0001:4202:15
Active State Snapshot
{
  "db_locked": true,
  "mem_usage": "142MB",
  "active_step": "eval_policy"
}
Fork & Fix

Branch reality, test the fix instantly.

Found the error? Fork the execution immediately before the crash. Inject your code fix or prompt adjustment, and the agent resumes exactly from that state. No setup required.

  • Hot-swap code during replay.
  • Shareable sandbox URLs for the team.
Dark Abstract Background

An agent booked 50 flights.
You have no idea why.

Traditional logs only tell you what happened. AgentTrace gives you the entire deterministic context to discover why, immediately.

production / worker-node-04
14:02:01.32 [info] Connecting to travel API...
14:02:01.45 [info] Received prompt: "Find flights to NYC"
14:02:02.10 [exec] Tool call triggered: book_flight(qty=50)
Fatal Error: Billing limit exceeded on API key
14:02:02.15 [warn] Agent loop forcefully terminated.
FAQs

Frequently Asked Questions

Everything you need to know about setting up and using AgentTrace.

Stop guessing.
Start debugging.

AgentTrace takes 2 minutes to install. Stop digging through print logs and start seeing exactly what your AI is doing.