Metadata-Version: 2.4
Name: agentcamp
Version: 0.1.0
Summary: AgentCamp SDK — where agents train their own experts. Observe, train, shadow, and activate cheaper expert models behind your existing agents.
Author: FintorAI
License-Expression: MIT
Project-URL: Homepage, https://agentcamp.ai
Project-URL: Repository, https://github.com/FintorAI/agentcamp
Project-URL: Changelog, https://github.com/FintorAI/agentcamp/blob/main/CHANGELOG.md
Keywords: agents,langgraph,llm,fine-tuning,agentcamp,distillation
Classifier: Development Status :: 3 - Alpha
Classifier: Intended Audience :: Developers
Classifier: Programming Language :: Python :: 3
Classifier: Programming Language :: Python :: 3.9
Classifier: Programming Language :: Python :: 3.10
Classifier: Programming Language :: Python :: 3.11
Classifier: Programming Language :: Python :: 3.12
Classifier: Typing :: Typed
Requires-Python: <4.0,>=3.9
Description-Content-Type: text/markdown
License-File: LICENSE
Provides-Extra: dev
Requires-Dist: pytest; extra == "dev"
Requires-Dist: pytest-cov; extra == "dev"
Requires-Dist: build; extra == "dev"
Requires-Dist: twine; extra == "dev"
Dynamic: license-file

# AgentCamp

[![PyPI version](https://img.shields.io/pypi/v/agentcamp.svg)](https://pypi.org/project/agentcamp/)
[![Python versions](https://img.shields.io/pypi/pyversions/agentcamp.svg)](https://pypi.org/project/agentcamp/)

**Where agents train their own experts. Same results, dramatically cheaper.**

AgentCamp observes your existing agents in production, trains a fine-tuned
expert from the traces, shadows it for safety, and then activates it behind
your stack — with automatic fallback to the teacher model.

> v0.1.0 is an early placeholder release. It ships a stable public client
> surface (`AgentCampClient`, `wrap_langgraph`, and `Mode`) so you can wire up
> integrations now. Calls are currently no-ops and will begin emitting traces
> once the training backend lands.

## Installation

```bash
pip install agentcamp
```

## Quick start

```python
import agentcamp

# Wrap an existing LangGraph graph so AgentCamp can observe its runs.
graph = agentcamp.wrap_langgraph(graph, project="support-agent")
```

Or use the client directly:

```python
from agentcamp import AgentCampClient, Mode

client = AgentCampClient(project="support-agent", mode=Mode.OBSERVE)
graph = client.wrap_langgraph(graph)
```

## Modes

| Mode       | What it does                                                        |
| ---------- | ------------------------------------------------------------------- |
| `OBSERVE`  | Collect traces from the teacher agent only.                         |
| `TRAIN`    | Build a fine-tuned expert from collected traces.                    |
| `SHADOW`   | Run the expert alongside the teacher without serving it.            |
| `ACTIVATE` | Serve the expert, falling back to the teacher when needed.          |

## Configuration

| Environment variable | Purpose                                  |
| -------------------- | ---------------------------------------- |
| `AGENTCAMP_API_KEY`  | API key used to authenticate the client. |
| `AGENTCAMP_BASE_URL` | Override the AgentCamp API base URL.     |

## Development

```bash
python -m venv .venv
source .venv/bin/activate
pip install -e '.[dev]'
pytest
```

## License

MIT — see [LICENSE](LICENSE).
