Metadata-Version: 2.4
Name: agentx-trace
Version: 0.1.0
Summary: Chrome DevTools for AI agents — capture, view, and debug agent traces
Author: SG1 Labs
License: MIT
Project-URL: Homepage, https://sg1-labs.us
Project-URL: Repository, https://github.com/gardnerscot/agentx
Keywords: ai,agent,debugging,llm,trace,observability
Classifier: Development Status :: 3 - Alpha
Classifier: Intended Audience :: Developers
Classifier: Topic :: Software Development :: Debuggers
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
Classifier: License :: OSI Approved :: MIT License
Classifier: Programming Language :: Python :: 3
Requires-Python: >=3.10
Description-Content-Type: text/markdown
Requires-Dist: fastapi>=0.100
Requires-Dist: uvicorn>=0.20

# AgentX 🔍

**Chrome DevTools for AI agents.** Capture, view, and debug your agent traces.

## Install

```bash
pip install agentx
```

## Usage

```python
from agentx import trace

@trace(output_dir="./traces")
def my_agent(prompt):
    response = call_llm(prompt)
    tool_output = call_tool("search", {"query": "something"})
    return response
```

Then view your traces:

```bash
agentx serve ./traces
```

Opens a browser at `http://localhost:8766` with a timeline view of every LLM call and tool call.

## Features

- 🎯 **@trace decorator** — one line, captures everything
- ⏱️ **Timeline view** — expandable entries with durations
- 🔢 **Token tracking** — prompt vs completion, per-call and totals
- 🔍 **Search & filter** — find specific calls by type or keyword
- 📊 **Stats sidebar** — steps, tokens, duration, LLM/tool breakdown
- 💾 **JSONL traces** — portable, version-controllable

## Coming Soon

- **Replay** — re-run traces against a different model
- **Step-through** — breakpoints and prompt injection
- **Diff view** — compare two runs side by side

---

Built by [SG1 Labs](https://sg1-labs.us) — small, useful tools. One-time purchase. No subscriptions.
