Local-first • MIT • 100% on your machine

Local-first
observability
for any AI agent.

Auto-detect and trace every token, tool call, and decision. One command. Pure SQLite. No cloud. No lock-in.

Get started in 30 seconds → View on GitHub
npm i -g @agenttrace-io/cli
or pip install agenttrace-io
Works with any agent — CLI, custom code, LangGraph, CrewAI, or fully hand-rolled.
Zero data leaves your machine by default
The challenge
Modern agents are getting more autonomous.
Most teams have almost no visibility into what they actually do.

Long-running agents produce complex chains of reasoning, tool calls, and sub-agent handoffs. Without structured traces you lose cost control and the ability to debug or audit what happened.

CAPABILITIES

Complete visibility.
Zero dependencies.

One lightweight approach that works for whatever agentic code you're running.

One-line CLI wrapper

Point it at any command and it auto-detects and records the full execution. No code changes required for CLI-based agents.

Precise cost & token tracking

Built-in pricing for common models plus support for custom rates. Per-agent, per-day, and per-model breakdowns with budget guardrails.

Multi-agent trace trees

See the full hierarchy when agents spawn sub-agents. Parent/child linking and execution graphs for complex systems.

Local dashboard + CLI

Run agenttrace dashboard for live stats, searchable runs, full trace drill-down, and export. Or stay in the terminal.

SDKs + framework middleware

Drop-in TypeScript and Python SDKs for custom code. Optional thin middleware for LangGraph, CrewAI and similar. The wrapper covers everything else.

SQLite + export escape hatches

Everything lives in one agenttrace.db file. Export JSON/CSV. Forward to OpenTelemetry. Webhooks for alerts. Full data ownership.

Three steps. No cloud. No drama.

01 — INSTALL
One command
npm i -g @agenttrace-io/cli
or pip install agenttrace-io
02 — TRACE
Wrap or instrument
Zero-config: agenttrace wrap your-agent ...
The wrapper auto-detects. Use the SDK or optional middleware where it makes sense.
03 — OBSERVE
Local dashboard + CLI
agenttrace dashboard or agenttrace runs
Everything stays in agenttrace.db.
THE LOCAL DASHBOARD

See everything at a glance.
Right on your machine.

Recent runs
Live updates
●
deep-research-run
2m ago • 47k tokens • $0.19
12 spans
●
autonomous-pipeline
14m ago • 128k tokens • $0.61
31 spans
●
customer-support-agent
1h ago • 19k tokens • $0.08
8 spans
autonomous-pipeline success
128,420 tokens • 4m 12s • $0.61
1
plan_research gpt-4o • 1.8k tokens
2
web_search tool • 420ms
3
read_page × 4 tools • 2.1s
4
synthesize_report claude-3-5-sonnet • 9.4k tokens
Full tree • 31 spans • Export JSON / CSV / OTLP

All data stays in agenttrace.db. Open source. Self-hostable. No telemetry.

Start tracing in under a minute.

No accounts. No credit card. No data leaves your machine.

TERMINAL
npm install -g @agenttrace-io/cli
TERMINAL
agenttrace wrap python research_agent.py
TERMINAL
agenttrace dashboard # opens at http://localhost:4317
Read the full quickstart & SDK docs →

Honest comparison

Built for teams that want the simplest possible local truth.

Capability AgentTrace Langfuse LangSmith
100% local / zero cloud ✓ Pure local SQLite Cloud or heavy self-host Cloud only
CLI zero-config wrapper ✓ First class No No
Single-file storage ✓ agenttrace.db ClickHouse / Postgres Cloud
Price for core tracing $0 forever (MIT) Usage + self-host cost Paid seats

Based on public documentation. AgentTrace focuses on simplicity and local ownership.

Ready to stop flying blind with your agents?
Install. Wrap. Understand.
Get started Star on GitHub
MIT licensed • Open source • Community contributions welcome