At a Glance:
Onyx is an open-source AI platform providing an application layer for LLMs, featuring agentic RAG, deep research, custom agents, and built-in code execution alongside multiple deployment modes and 50+ connectors.
Overview:
Onyx is an open-source application layer for large language models that provides a feature-rich interface designed to be self-hosted. The platform integrates multiple LLM capabilities including retrieval-augmented generation, web search, code execution, and document generation. It connects to external data sources through over 50 indexing-based connectors or MCP. Onyx supports all major LLM providers, both self-hosted options like Ollama and LiteLLM and proprietary services like Anthropic and OpenAI, and offers two distinct deployment modes to accommodate different resource requirements and use cases.
Key Decision Points:
Deployment flexibility: Offers both a Lite mode requiring under 1GB memory for quick testing or chat-focused use, and a Standard mode with full RAG indexing, background workers, and performance optimizations via Redis and MinIO.
Deployment platforms: Supports Docker, Kubernetes, Helm/Terraform with guides for major cloud providers, plus a cloud-hosted option without self-deployment.
LLM provider flexibility: Supports all major LLM providers including self-hosted options (Ollama, LiteLLM, vLLM) and proprietary services (Anthropic, OpenAI, Gemini).
Integration scope: Provides over 50 indexing-based connectors out of the box and supports MCP for connecting agents to external applications.
Enterprise features: Community Edition under MIT license covers core Chat, RAG, Agents, and Actions, while Enterprise Edition adds SSO, RBAC, analytics, query history, and whitelabeling.
Core Features:
Agentic RAG: Hybrid index combined with AI agents for information retrieval providing search and answer generation.
Deep Research: Multi-step research flow producing in-depth reports, ranked top of leaderboard as of February 2026.
Custom Agents: Build AI agents with unique instructions, knowledge bases, and configurable actions.
Code Execution: Execute code in a sandbox for data analysis, graph rendering, or file modification.
Voice Mode: Text-to-speech and speech-to-text chat functionality.
Image Generation: Generate images based on user prompts.
Use Cases:
Developers and teams who want to self-host an AI platform with RAG, agents, and web search capabilities using their preferred LLM providers.
Users who need to connect LLMs to multiple data sources through the 50+ built-in connectors or MCP integrations.
Teams requiring lightweight deployment for chat and agent functionalities through the Onyx Lite mode without running the full indexing stack.
Organizations evaluating enterprise-ready AI platforms with SSO, RBAC, and usage analytics before committing to the Enterprise Edition.
Open-Source Alternative Value:
Onyx provides an open-source, self-hostable alternative to proprietary AI platforms by offering a complete application layer for LLMs under the MIT license for its Community Edition. Users can deploy the platform through Docker, Kubernetes, or major cloud providers, support their choice of LLM providers including self-hosted models via Ollama or vLLM, and access features like agentic RAG, deep research, custom agents, and code execution. The project also provides 50+ connectors and MCP support for integration with external applications without depending on a hosted service.
