Metadata-Version: 2.4
Name: sythra
Version: 0.1.0
Summary: Runtime intelligence for AI agents: find the first point where a run went wrong, what it cost, and how to fix it.
Author-email: Sythra <sythra.hq@gmail.com>
Project-URL: Homepage, https://sythra.vercel.app
Project-URL: Repository, https://github.com/sythrahq/sythra
Keywords: ai,agents,observability,llm,debugging
Classifier: Programming Language :: Python :: 3
Classifier: Operating System :: OS Independent
Requires-Python: >=3.10
Description-Content-Type: text/markdown
Requires-Dist: httpx>=0.24

# Sythra

Runtime intelligence for AI agents. When a run goes wrong, Sythra finds the first point where it diverged, shows what it cascaded into, what it cost, and what to do about it.

## Install

```
pip install sythra
```

## Quick start

Create an agent in your dashboard at https://sythra.vercel.app to get an API key, then wrap your agent function:

```python
from sythra import Sythra

sythra = Sythra(api_key="YOUR_API_KEY", agent_name="my-agent")

@sythra.observe
def run_agent(question):
    ...  # your agent code
```

Each call is recorded with its outcome, duration and steps. Works with LangChain, LangGraph, CrewAI and the OpenAI Agents SDK.

## Privacy

Sythra records what your agent did (tools, models, cost, outcome), not what it said. Prompt and response content is never captured unless you turn it on.

## Links

- Dashboard: https://sythra.vercel.app
- Source: https://github.com/sythrahq/sythra
- Support: sythra.hq@gmail.com
