ABI

Agentic Brain Infrastructure

Multi-agent AI Operating System with semantic knowledge graphs, ontology-driven reasoning, and intelligent workflow automation.

⭐ Star and follow to stay updated!

Quick Start

Prerequisites

# Install uv package manager
curl -LsSf https://astral.sh/uv/install.sh | sh

# Install Docker Desktop (for local services)
# https://www.docker.com/products/docker-desktop

Local Development

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

# Install dependencies (Python + frontend)
uv sync --all-extras

# Install frontend dependencies
cd libs/naas-abi/naas_abi/apps/nexus/apps/web
pnpm install
cd ../../../../..

# Create local config
cp config.yaml.example config.yaml
# Edit config.yaml with your API keys

# Configure for local development (update .env)
# Change Docker hostnames to localhost:
#   POSTGRES_HOST=localhost (not postgres)
#   QDRANT_HOST=localhost (not qdrant)
#   MINIO_HOST=localhost (not minio)

# Start infrastructure
docker compose up -d postgres fuseki rabbitmq

# Start platform
uv run abi stack start

Platform will launch at:

• 🌐 Nexus UI: http://localhost:3000
• 📊 Nexus API: http://localhost:9879
• 🤖 Agent API: http://localhost:8001
• 🗄️ Fuseki: http://localhost:3030

CLI Commands

uv run abi stack start         # Start all services
uv run abi stack stop          # Stop all services
uv run abi stack status        # Show service health
uv run abi stack logs [svc]    # Stream BFO logs (api|web|core|all)
uv run abi seed-jena           # Populate graph database
uv run abi chat                # Interactive agent chat

Configuration

Minimal config (loads AbiAgent only):

modules:
  - module: naas_abi
    enabled: true
  - module: naas_abi_core.modules.templatablesparqlquery
    enabled: true
  - module: naas_abi_marketplace.ai.chatgpt
    enabled: true

services:
  triple_store:
    triple_store_adapter:
      adapter: "apache_jena_tdb2"
      config:
        jena_tdb2_url: "http://admin:abi@localhost:3030/ds"

Architecture

Four-Layer AI Operating System:

1. User Layer: Chat UI, REST API, MCP Protocol
2. Agent Layer: ABI SuperAssistant + 20+ domain experts
3. Storage Layer: Knowledge Graph (Jena/Oxigraph), Vector DB (Qdrant), Memory (PostgreSQL)
4. Execution Layer: Ontologies (BFO), Workflows, Integrations, Analytics

graph LR
    USER[👤 User] --> APPS[📱 Apps]
    APPS --> AGENTS[🧠 Agents]
    AGENTS --> STORAGE[(💾 Storage)]
    AGENTS --> EXEC[⚙️ Components]
    STORAGE --> KG[(Knowledge Graph)]
    STORAGE --> VDB[(Vector DB)]
    EXEC --> ONT[Ontologies]
    EXEC --> WF[Workflows]

Loading

Key Features

🤖 Multi-Model AI

• Cloud: ChatGPT, Claude, Gemini, Grok, Llama, Mistral
• Local: Qwen, DeepSeek, Gemma (via Ollama)
• Supervisor: ABI agent with intelligent routing

🧠 Knowledge Management

• Semantic Graph: BFO-compliant RDF ontologies
• SPARQL Queries: 30+ optimized queries
• Vector Search: Intent matching via embeddings
• Memory: Persistent conversation context

🏪 Marketplace

• Domain Experts: 20+ agents (Engineer, Analyst, Creator, etc.)
• Integrations: GitHub, LinkedIn, Google, PostgreSQL, ArXiv, etc.
• Modular: Enable/disable via config.yaml

⚙️ Automation

• Workflows: End-to-end process automation
• Pipelines: Data → Semantic transformation
• Event-Driven: Knowledge graph triggers
• Integrations: External APIs and exports

🌐 Multiple Interfaces

• Terminal: uv run abi chat - Interactive CLI
• REST API: HTTP endpoints
• MCP Protocol: Claude Desktop / VS Code
• Web UI: http://localhost:3000

Production Deployment

Deploy to Naas Cloud

uv run abi deploy naas

Requires Naas subscription. Deploys as containerized API at https://{space}.default.space.naas.ai

Self-Hosted Docker

docker-compose up -d

Full stack with PostgreSQL, Fuseki, Qdrant, MinIO

Services

Service	Port	Purpose
Nexus Web	3000	Frontend UI
Nexus API	9879	Platform API
Agent API	8001	Agent execution
Fuseki	3030	Graph database
PostgreSQL	5432	Relational DB
Qdrant	6333	Vector DB
MinIO	9000/9001	Object storage

Why ABI?

Ontology-Based AI for Freedom to Reason: When semantic alignment meets kinetic action through ontology-driven systems, we get one of the most powerful technologies ever created. This power should be distributed, not concentrated - the ability to understand, reason, and act upon complex information is fundamental to human autonomy and democratic society.

Built on International Standards:

• ISO/IEC 42001:2023 - AI Management Systems
• ISO/IEC 21838-2:2021 - Basic Formal Ontology (BFO)
• EU AI Act compliance-ready

For:

• 👤 Individuals: Run locally, own your data
• ⚡ Pro: Automate workflows, optimize costs
• 👥 Teams: Share knowledge, build agents
• 🏢 Enterprise: Deploy at scale, full control

Research & Development

Collaborative effort between:

• NaasAI - Applied AI Research Lab
• OpenTeams - Open SaaS Infrastructure
• University at Buffalo - Research University
• NCOR - National Center for Ontological Research
• Forvis Mazars - Global Audit & Consulting

Requirements

System:

• Python 3.10+
• uv package manager
• Git

Hardware (Minimal - Cloud AI):

• 2GB+ RAM
• 500MB disk

Hardware (Full - Local/Docker):

• 8GB+ RAM
• 10GB+ disk
• Docker Desktop

API Keys (at least one):

• OpenAI, Anthropic, Google AI, OpenRouter, or other LLM providers

Contributing

We welcome contributions! See CONTRIBUTING.md for guidelines.

License

MIT License - see LICENSE

For enterprise support: support@naas.ai