Onyx is a self-hosted AI platform that combines enterprise search, RAG (Retrieval-Augmented Generation), chat, and custom agents. The project was created to address the need for organizations to run AI assistants on their own infrastructure, keeping data, prompts, and logs within their security boundary.
Onyx emerged from the growing demand for private, controllable AI deployments in the enterprise sector. As organizations began experimenting with LLMs, they faced a critical challenge: how to leverage AI capabilities while maintaining data sovereignty and compliance requirements. Onyx positioned itself as a solution that combines:
- Enterprise Search - Connecting to 40+ data sources (Slack, Confluence, Google Drive, SharePoint, etc.)
- RAG Pipeline - Hybrid search with knowledge graphs for accurate retrieval
- Chat Interface - User-friendly interface for interacting with AI
- Custom Agents - Build AI agents with specific instructions and tool access
Like many GenAI platforms, Onyx evolved through several phases:
¶ Phase 1: Core Chat and Search
Initial releases focused on providing a functional chat interface with basic document search capabilities. The early architecture established the foundation:
- FastAPI backend for API services
- Next.js frontend for the user interface
- PostgreSQL for structured data storage
- Vespa for vector search and semantic retrieval
As adoption grew, Onyx expanded its connector ecosystem to integrate with enterprise systems:
- Google Workspace (Drive, Docs, Gmail)
- Microsoft 365 (SharePoint, OneDrive, Teams)
- Atlassian (Confluence, Jira)
- Slack and Discord
- GitHub and GitLab
- Salesforce and HubSpot
¶ Phase 3: Agents and Actions
The platform evolved to support custom AI agents with:
- Configurable instructions and knowledge bases
- Tool integration via MCP (Model Context Protocol)
- Code interpreter for data analysis
- Web search and browsing capabilities
- Deep research with multi-step agentic workflows
Onyx introduced a dual-license model:
- Community Edition (CE) - MIT licensed, fully functional self-hosted platform
- Enterprise Edition (EE) - Additional features for larger organizations including advanced RBAC, audit logging, and priority support
Onyx’s deployment options matured to support various infrastructure needs:
| Deployment Method |
Use Case |
| Docker Compose |
Small to medium deployments, quick setup |
| Kubernetes |
Large teams, high availability requirements |
| Terraform |
Infrastructure-as-Code workflows |
| Cloud Providers |
AWS EKS, Azure VMs, GCP GKE deployments |
| Air-gapped |
Fully isolated environments with self-hosted LLMs |
The onyx-dot-app organization maintains the project as an open-source initiative:
- Primary repository: https://github.com/onyx-dot-app/onyx
- Documentation: https://docs.onyx.app
- Community contributions drive connector development and deployment improvements
- Regular releases with new features and security updates
Onyx is built with modern, production-ready technologies:
| Layer |
Technology |
| Backend |
Python 3.11+ with FastAPI |
| Frontend |
Next.js 16+ with TypeScript |
| Database |
PostgreSQL 15.2 |
| Vector Search |
Vespa 8.x |
| Cache |
Redis 7.x |
| Task Queue |
Celery 5.x |
| Web Server |
Nginx |
Today, Onyx is positioned as a comprehensive enterprise AI platform that:
- Scales to tens of millions of documents
- Supports 40+ enterprise connectors
- Enables custom agent creation with tool access
- Provides collaboration features (chat sharing, feedback, analytics)
- Maintains security with SSO (OIDC/SAML/OAuth2), RBAC, and credential encryption
- Runs in air-gapped environments with self-hosted LLMs (Ollama, vLLM)
The project continues to evolve with focus on:
- Expanded connector ecosystem
- Improved agent capabilities and tool integration
- Enhanced security and compliance features
- Better support for self-hosted and open-source LLMs
- Streamlined deployment and operations