Onyx

Onyx is an open-source AI platform that consolidates internal knowledge management, secure data retrieval, and AI agent deployment in a single chat interface. Built for organizations handling sensitive data, it provides grounded, source-backed answers from private documents and databases.

RAG, 50+ Connectors, and Agentic Deep Research

Onyx provides advanced capabilities for LLMs, focusing on Retrieval-Augmented Generation (RAG), web search, code execution, file creation, and deep research. It allows users to build custom AI agents and integrates with over 50 indexing-based connectors to various data sources. The platform aims to deliver reliable and grounded answers by combining hybrid search, advanced RAG, contextual retrieval, and LLM-based knowledge graphs.

Enterprise Knowledge Search and Air-Gapped Deployments

Onyx targets teams and organizations, particularly those with security and data privacy concerns, looking to deploy AI internally. Specific use cases include:

• Enterprise Knowledge Search: Q&A with fine-grained access control across tools like Google Drive, Notion, GitHub, Slack, and SharePoint.
• Secure Deployments: Offline or air-gapped deployments for highly secure environments such as defense, healthcare, and finance.
• Custom AI Agents: Building tailored agents and deep research assistants.

Its emphasis on self-hosting, air-gapped deployments, and enterprise security features makes it highly suitable for regulated industries. The platform’s "Agentic RAG" and "Deep Research" capabilities, involving multi-step research flows and AI agents, are designed for superior search and answer quality in complex problem-solving.

Docker, Kubernetes, and 50+ Data Connectors

Onyx is highly configurable and offers significant technical flexibility:

• Deployment Options: Self-hostable via Docker, Kubernetes, or Terraform, supporting fully air-gapped deployments.
• LLM Compatibility: Works with major LLM providers, both self-hosted (e.g., Ollama, Llama) and proprietary (e.g., Anthropic, OpenAI, Gemini). Users can bring their own API keys on Enterprise plans.
• Data Connectors: Over 50 indexing-based connectors for applications like Google Drive, Salesforce, and SharePoint, featuring automated syncing and permission-aware search.
• RAG Architecture: Utilizes hybrid search, advanced RAG, contextual retrieval, and LLM-based knowledge graphs, including dense, sparse, and hybrid retrieval strategies.
• Security Features: Includes SSO, RBAC, credential encryption, audit logs, and permission-aware retrieval.
• Additional Capabilities: Features web search, sandboxed code execution, image generation, and voice mode.

Internal benchmarks from early 2026 reported a 64-76% win rate against competitors like ChatGPT Enterprise and Claude Enterprise for workplace questions.

Free Community Edition, Paid Cloud Plans, and Enterprise

Onyx offers several options to suit different organizational needs:

Premium LLM models are billed separately based on token usage. A 14-day free trial is available for Onyx Cloud.

RAG Reliability, Setup Complexity, and Token Costs

Onyx has noteworthy trade-offs:

• RAG reliability: Some users report that Onyx’s RAG functionality can be unreliable, occasionally failing to retrieve or effectively use document context, leading to generic answers rather than source-backed responses.
• General RAG challenges: Like all RAG systems, it faces issues with outdated knowledge bases and chunking problems that can reduce answer quality.
• Setup complexity: Self-hosting with Docker, Kubernetes, or Terraform requires DevOps knowledge, and configuring the 50+ data connectors can be time-consuming.
• Token costs: Premium LLM models are billed separately based on usage, which can add up for teams processing large document sets.