Quickstart

Get a running ABI stack in under 10 minutes.

---

Prerequisites

• Docker Desktop - required for local services
• uv - Python package manager
• An OpenRouter API key - optional but recommended for cloud models. Skip for local-only mode.

---

1) Clone and install

git clone https://github.com/jupyter-naas/abi.git
cd abi

# Install all dependencies
uv sync --all-extras
uv pip install greenlet

---

2) Configure

# Copy the example config
cp config.yaml.example config.yaml
cp .env.example .env

Edit .env and set at minimum:

# Cloud mode (recommended for full capability)
OPENROUTER_API_KEY=sk-or-...

# Local mode (no internet, no API key required)
# Set ai_mode: "local" in config.yaml instead, and install Ollama

---

3) Start the stack

abi stack start

This starts all Docker services (Fuseki, PostgreSQL, RabbitMQ, MinIO) and confirms readiness before returning. Check status at any time:

abi stack status

---

4) Chat with ABI

abi chat

This launches the terminal agent. Type a question - ABI will route it to the correct specialized agent.

Or start the API server:

uv run python -m naas_abi_core.apps.api.api

Then open:

• http://localhost:9879/ - Nexus web app
• http://localhost:9879/docs - API documentation (Swagger)
• http://localhost:9879/redoc - API documentation (Redoc)

---

AI mode: cloud vs. local

Cloud mode (default): uses OpenRouter to access GPT-4o, Claude 3.5 Sonnet, Gemini 2.0, and others. Requires OPENROUTER_API_KEY.

Local mode: all AI processing runs on your machine via Ollama. No data leaves your network.

# Install Ollama: https://ollama.com/download
ollama pull qwen3:8b   # or deepseek-r1:8b or gemma3:4b

Set in config.yaml:

global_config:
  ai_mode: "local"

---

Useful commands

# Service management
abi stack start          # Start all services
abi stack stop           # Stop all services
abi stack status         # Health check
abi stack logs api       # Stream API logs

# Development
make api                 # Start API server (port 9879)
make mcp                 # Start MCP server for Claude Desktop
make sparql-terminal     # Interactive SPARQL query terminal

# Data
make datastore-pull      # Pull datastore from remote
make datastore-push      # Push local changes to remote

# Testing
make test                # Run all tests
make check               # Code quality checks

---

Service URLs (when running)

Service	URL	Purpose
ABI API	http://localhost:9879	REST API + Nexus web
Fuseki SPARQL	http://localhost:3030	Knowledge graph admin
Dagster	http://localhost:3001	Orchestration UI
MCP server	http://localhost:8000	Model Context Protocol
Qdrant	http://localhost:6333	Vector store admin

Next: Configuration Reference or Build your first module.