# Webber

Multi-Agent AI Development System - a FastAPI-based service that orchestrates local LLM agents for code exploration, planning, and task execution.

## Overview

Webber provides autonomous AI agents similar to Claude Code but running locally with configurable models via Ollama. It's designed for:

- **Explore Agent** - Fast codebase navigation and code search
- **Plan Agent** - Implementation design and step-by-step planning
- **Task Agent** - Autonomous multi-step code generation and modification

Built on [PydanticAI](https://ai.pydantic.dev/) for structured LLM interactions.

## Quick Start

### Prerequisites

- Python 3.12+
- [Ollama](https://ollama.ai/) with models installed
- (Optional) Tatlock for multi-tenant authentication

### Installation

```bash
# Clone the repository
git clone https://git.schweitz.internal/jpmschweitzer/webber.git
cd webber

# Create virtual environment
python -m venv .venv
source .venv/bin/activate

# Install dependencies
pip install -r requirements.txt

# For development (includes testing and linting tools)
pip install -r requirements-dev.txt
```

### Configuration

```bash
# Copy example config
cp .env.example .env

# Edit .env with your settings
# At minimum, configure OLLAMA_URL to point to your Ollama instance
```

### Running

```bash
# Development (with auto-reload)
./wakeup.sh

# Or manually
uvicorn src.main:app --host 0.0.0.0 --port 8086 --reload
```

The service will be available at `http://localhost:8086`. API docs at `/docs`.

## Configuration

All settings via environment variables or `.env` file:

| Variable | Default | Description |
|----------|---------|-------------|
| `DEBUG` | `false` | Enable debug mode |
| `LOG_LEVEL` | `INFO` | Logging level |
| `PORT` | `8086` | Server port |
| `OLLAMA_URL` | `http://192.168.86.149:11434` | Ollama API URL |
| `OLLAMA_AGENT_MODEL` | `mistral-nemo-large:latest` | Model for agent reasoning |
| `OLLAMA_EMBED_MODEL` | `nomic-embed-text:latest` | Model for embeddings |
| `TOOL_TIMEOUT_SECONDS` | `120` | Tool execution timeout |
| `SANDBOX_ENABLED` | `true` | Sandbox tool execution |
| `ALLOWED_PATHS` | `[]` | Paths accessible to tools |

See [.env.example](.env.example) for full configuration options.

## Development

### Code Quality

```bash
# Type checking
mypy src/

# Linting
ruff check src/ tests/

# Auto-fix lint issues
ruff check src/ tests/ --fix

# Format code
ruff format src/ tests/
```

### Testing

```bash
# Run all tests
pytest tests/ -v

# With coverage
pytest tests/ --cov=src --cov-report=html
```

### Security Audit

```bash
# Check dependencies for CVEs
pip-audit
```

## Architecture

Webber uses a domain-based architecture with clean separation of concerns:

```
src/
├── main.py              # FastAPI app entry point
├── shared/              # Cross-cutting infrastructure
│   ├── base.py          # BaseController, BaseSchema
│   ├── config.py        # Settings from pyproject.toml + env
│   ├── logging.py       # @logged decorator with timing
│   └── exceptions.py    # Exception hierarchy
└── domains/             # Feature domains
    ├── health/          # Health check endpoints
    ├── agents/          # Agent orchestration
    └── tools/           # Tool execution (file, shell, search)
```

See [docs/architecture.md](docs/architecture.md) for detailed patterns and conventions.

## API Endpoints

| Endpoint | Method | Description |
|----------|--------|-------------|
| `/` | GET | Service information |
| `/health` | GET | Health check for monitoring |
| `/docs` | GET | Interactive API documentation |

## Docker

```bash
# Build
docker build -t webber .

# Run
docker run -p 8086:8086 --env-file .env webber
```

The container includes a healthcheck that pings `/health` every 30 seconds.

## Deployment

Deployed via Gitea Actions CI/CD:

1. Tag a release (`git tag v0.x.x && git push --tags`)
2. Workflow builds and pushes Docker image
3. Watchtower auto-deploys to production

Production runs in Portainer `agents` stack on the `docker-dataplane` network.

## Status

**Alpha** - Core infrastructure is complete. Agent and tool implementations are in progress.

## License

MIT