Onyx is a feature-rich, self-hosted AI platform that combines enterprise search, RAG (Retrieval-Augmented Generation), chat, and custom agents. It is designed for organizations that need a private AI assistant with controlled access to internal knowledge sources. Onyx supports 40+ enterprise connectors, hybrid search with knowledge graphs, deep research capabilities, and can scale to tens of millions of documents. It runs in air-gapped environments and integrates with any LLM provider (OpenAI, Anthropic, Gemini, Ollama, vLLM).
- 🤖 Custom Agents - Build AI agents with unique instructions, knowledge, and actions
- 🌍 Web Search - Browse with Google PSE, Exa, Serper, Firecrawl, or in-house scraper
- 🔍 RAG - Hybrid search + knowledge graph for uploaded files and ingested documents
- 🔄 Connectors - 40+ application integrations (Slack, Confluence, Google Drive, SharePoint, etc.)
- 🔬 Deep Research - Agentic multi-step search for in-depth answers
- ▶️ Actions & MCP - Interact with external systems via AI agents
- 💻 Code Interpreter - Execute code, analyze data, render graphs, create files
- 🎨 Image Generation - Generate images from prompts
- 👥 Collaboration - Chat sharing, feedback, user management, analytics
- 🔐 Security - SSO (OIDC/SAML/OAuth2), RBAC, credential encryption
- 📄 Document Permissioning - Mirrors user access from external applications
- 🚀 Enterprise Search - Scales to tens of millions of documents
- Enterprise knowledge assistant with private data
- Internal document search across multiple systems
- Private AI chat for teams with SSO integration
- Air-gapped deployments with self-hosted LLMs
- Customer-facing AI assistants with access controls
- Automated research and analysis workflows
| Component |
Technology |
| Backend |
Python 3.11+ with FastAPI 0.133.1 |
| Frontend |
Next.js 16.1.6 (React 19.2.4) with TypeScript |
| Primary Database |
PostgreSQL 15.2 |
| Vector Search |
Vespa 8.609.39 |
| Cache |
Redis 7.4 |
| Task Queue |
Celery 5.5.1 |
| Web Server |
Nginx 1.25.5 |
| Optional Search |
OpenSearch 3.4.0 |
| Object Storage |
MinIO (S3-compatible) |
| Containerization |
Docker, Kubernetes |
| Resource |
Minimum |
Preferred |
| CPU |
4 vCPU |
8+ vCPU |
| RAM |
10 GB |
16+ GB |
| Disk |
32 GB + ~2.5x indexed data |
500 GB (for orgs <5000 users) |
Note: Vespa does not allow writes once disk usage hits 75%.
- Community Edition (CE): MIT License (open-source, self-hosted)
- Enterprise Edition (EE): Proprietary license (additional features for larger organizations)
- Actively maintained open-source project
- Dual-license model (CE + EE)
- Regular releases with new connectors and features
- Strong enterprise adoption for private AI deployments
¶ History and References
Any questions?
Feel free to contact us. Find all contact information on our contact page.