FlowTrace

Visualizes and debugs multi-step LLM agent workflows with interactive graphs and local inspection

$pip install flowtrace

Why FlowTrace?

AI agents chain dozens of LLM calls, tool invocations and memory reads yet failures are opaque. Existing observability is cloud-only, framework-specific or noisy. FlowTrace is a lightweight local dashboard that auto-instruments any Python agent code, rendering clean interactive graphs so developers can instantly understand and fix behavior.

Features

Everything you need, nothing you don't.

πŸ“

Zero-config tracing

Context manager and @trace decorator capture steps with one import and no framework lock-in.

πŸ•ΈοΈ

Interactive graphs

SVG node-link diagram shows execution flow, nesting, timing and step types using a strict grid layout.

πŸ”Ž

Prompt inspector

Click any node to view raw prompts, responses, token counts, latency and metadata in side panels.

πŸ› 

Tool call browser

Lists every tool invocation with input, output and the exact code location that triggered it.

πŸ’Ύ

Local persistence

Traces saved to SQLite; browse history, compare runs and filter by duration or success.

▢️

Trace replay

Re-execute a saved trace with mocked tools to experiment without hitting real APIs.

πŸ“‘

HTTP ingestion

Simple POST endpoint accepts traces from non-Python agents or remote services.

πŸ“Š

Summary stats

Aggregate view shows token usage, success rate, slowest steps and common failure patterns.

See it in action

From install to result in seconds.

FlowTrace terminal demo

Get started

Three steps. Full docs live in the README.

Install
pip install flowtrace
Run
flowtrace serve
Test
pytest

Like it? Give it a ⭐

Stars help other developers discover FlowTrace. Issues and PRs are very welcome.

Star on GitHub