https://github.com/gradrix/dendrite
🧠 Self-organizing AI agent with dendrite spawning - decomposes complex goals into neurons, spawns parallel sub-tasks, and aggregates results. Features smart parameter extraction, context isolation, and automatic data compaction. Built for Strava analytics with extensible tool system.
https://github.com/gradrix/dendrite
agentic-ai ai-agent autonomous-agent llm mistral neural-architecture ollama python strava strava-api task-decomposition
Last synced: 3 months ago
JSON representation
🧠 Self-organizing AI agent with dendrite spawning - decomposes complex goals into neurons, spawns parallel sub-tasks, and aggregates results. Features smart parameter extraction, context isolation, and automatic data compaction. Built for Strava analytics with extensible tool system.
- Host: GitHub
- URL: https://github.com/gradrix/dendrite
- Owner: gradrix
- Created: 2025-10-23T21:25:12.000Z (8 months ago)
- Default Branch: main
- Last Pushed: 2025-12-21T20:35:40.000Z (6 months ago)
- Last Synced: 2025-12-23T08:51:24.812Z (6 months ago)
- Topics: agentic-ai, ai-agent, autonomous-agent, llm, mistral, neural-architecture, ollama, python, strava, strava-api, task-decomposition
- Language: Python
- Homepage:
- Size: 1.4 MB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- Roadmap: ROADMAP.md
Awesome Lists containing this project
README
# Dendrite
A self-contained AI neural engine with tool execution, scheduled goals, and 100% local LLM inference via llama.cpp.
## Features
- **Local LLM**: Built-in llama.cpp server (Mistral 7B) - no external APIs
- **Tool System**: Semantic tool discovery and execution (Strava, memory, calculator, etc.)
- **Scheduler**: Run goals on cron schedules or intervals
- **PostgreSQL Storage**: Persistent tool data and credentials
- **GPU Acceleration**: Auto-detects NVIDIA GPU for faster inference
## Quick Start
```bash
# Clone
git clone https://github.com/gradrix/dendrite.git
cd dendrite
# Start (auto-detects GPU)
./start.sh
# Run a single goal
./start.sh goal "What is 2+2?"
# Run scheduler daemon
./start.sh scheduler
```
## Commands
```bash
./start.sh # Start services (auto-detect GPU)
./start.sh goal "..." # Run single goal and exit
./start.sh scheduler # Run scheduler daemon (uses goals.yaml)
./start.sh api # Start HTTP API server
./start.sh stop # Stop all services
./start.sh status # Show service status
./start.sh logs # Follow logs
./start.sh test # Run tests
./start.sh help # Show help
```
## Configuration
### Environment (.env)
```bash
# RAM profile for model selection (8gb, 16gb, 32gb)
RAM_PROFILE=32gb
# GPU VRAM (auto-detected if not set)
VRAM_GB=32
# Strava OAuth (optional)
STRAVA_CLIENT_ID=your_id
STRAVA_CLIENT_SECRET=your_secret
```
### Scheduled Goals (goals.yaml)
```yaml
goals:
- id: collect_kudos
goal: "Use strava_collect_kudos_givers with hours_back=48"
schedule: cron
cron: "0 */4 * * *" # Every 4 hours
enabled: true
- id: reciprocate_kudos
goal: "Use strava_reciprocate_kudos with count=30 and max_age_hours=24"
schedule: cron
cron: "0 */6 * * *" # Every 6 hours
enabled: true
settings:
check_interval: 60
```
## Architecture
```
┌─────────────────────────────────────────────────────────┐
│ Orchestrator │
│ Routes goals to appropriate neurons based on intent │
└─────────────────────────────────────────────────────────┘
│
┌─────────────────┼─────────────────┐
▼ ▼ ▼
┌───────────┐ ┌───────────┐ ┌───────────┐
│ Intent │ │ Tool │ │ Generative│
│ Neuron │ │ Neuron │ │ Neuron │
└───────────┘ └───────────┘ └───────────┘
│ │ │
▼ ▼ ▼
┌───────────┐ ┌───────────┐ ┌───────────┐
│ LLM │ │ Tool │ │ LLM │
│ Client │ │ Registry │ │ Client │
└───────────┘ └───────────┘ └───────────┘
│
▼
┌─────────────────────────┐
│ Tools (Strava, etc) │
│ PostgreSQL Storage │
└─────────────────────────┘
```
## Tools
### Built-in
- `calculator` - Math expressions
- `memory_write` / `memory_read` - Persistent key-value storage
- `current_datetime` - Current date/time
### Strava Integration
- `strava_get_activities` - Get your activities
- `strava_get_dashboard_feed` - Get friends' activities
- `strava_give_kudos` - Give kudos to an activity
- `strava_collect_kudos_givers` - Track who gives you kudos
- `strava_reciprocate_kudos` - Auto-kudos back to givers
- `strava_list_kudos_givers` - List known kudos givers
## Services
| Service | Port | Description |
|---------|------|-------------|
| llama-gpu | 8080 | llama.cpp server (Mistral 7B) |
| postgres | 5432 | PostgreSQL with pgvector |
| redis | 6379 | Message bus / caching |
## Development
```bash
# Run tests
./start.sh test
# Run specific tests
./start.sh test -k "test_strava"
# Access container shell
./scripts/docker/shell.sh
```
## Project Structure
```
├── main.py # Entry point
├── start.sh # Main startup script
├── goals.yaml # Scheduled goals config
├── docker-compose.yml # Services definition
├── neural_engine/
│ └── v2/
│ ├── core/ # Config, LLM, Orchestrator
│ ├── neurons/ # Intent, Tool, Generative, Memory
│ ├── tools/ # Tool implementations
│ ├── scheduler/ # Goal scheduling
│ ├── forge/ # Dynamic tool creation
│ ├── cli.py # Command-line interface
│ └── api.py # HTTP API
└── scripts/
├── db/ # Database migrations
├── docker/ # Docker helper scripts
└── testing/ # Test runners
```
## License
MIT