dvlin-dev

dvlin-dev / agent-trace

Public

Desktop app for tracing Claude Code API traffic, sessions, tools, and thinking blocks

13
2
100% credibility
Found Mar 17, 2026 at 13 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
TypeScript
AI Summary

Agent Trace is a desktop app that captures, organizes, and displays real-time communication traces between AI agent clients and providers for inspection and debugging.

How It Works

1
🔍 Discover Agent Trace

You find this handy desktop tool that lets you peek inside conversations between your AI helpers and their services.

2
📥 Install and launch

Download the app, install it like any other program, and open it up to get started.

3
⚙️ Pick your AI service

Choose the AI provider you're using, like Claude or Codex, and name your setup.

4
🔌 Connect everything

Tell the app where to forward chats and set a local spot for your AI tool to send messages through.

5
▶️ Start listening

Flip the switch to begin capturing your AI's back-and-forth talks in real time.

6
👀 Watch traces appear

Run your AI agent normally and see structured conversations, tools, and thoughts pop up instantly.

✅ Inspect and debug

Dive into timelines, spot issues, and understand exactly how your AI is thinking and acting.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 13 to 13 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is agent-trace?

Agent Trace is a TypeScript desktop app that proxies API calls from tools like Claude Code and Codex CLI, capturing requests, responses, sessions, tools, and reasoning blocks for real-time debugging. It normalizes traffic into searchable timelines with context chips for injected prompts, token counts, and structured views, all stored locally in SQLite—no cloud needed. Developers point their agent client to a local port via env vars like `ANTHROPIC_BASE_URL=http://127.0.0.1:8888`, and get instant agent trace visualization in a clean Electron/React UI.

Why is it gaining traction?

It beats generic tracers like Datadog trace agent github tools by focusing on AI agent protocols: auto-groups sessions via metadata or message supersets, highlights thinking blocks and tool schemas, and collapses system noise into labeled chips. Multi-profile support lets you run Anthropic and OpenAI side-by-side without config fights, delivering agent trace evaluation in seconds. The local-only design and real-time inspector make debugging agent flows feel effortless.

Who should use this?

AI agent builders tweaking Claude Code or Codex CLI who hit opaque tool failures or reasoning loops. Ideal for solo devs or small teams doing agent trace bedrock tests, visualizing multi-turn chats in desktop app chatgpt setups, or auditing prompt injections before production.

Verdict

Worth trying for agent trace github fans prototyping agents—features punch above its 13 stars—but the 1.0% credibility score flags early maturity with thin docs and unproven scale. Solid personal tool; monitor for wider adoption.

(187 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.