Onyx

Key features

• More than 50 connected knowledge sources. Index Slack, Google Drive, Confluence, Jira, GitHub, Salesforce, and dozens more with automated syncing, so the AI always works from current company knowledge instead of manually uploaded files.[1][2][3]

• Permission-aware retrieval. Source permissions sync into the index and are enforced at query time, so each person only sees documents they already have access to, whether asking in chat, search, or through a bot.[1][2]

• Agentic RAG over a hybrid index. A custom agent harness tuned for retrieval combines vector and keyword indexes for answers the vendor's open benchmark places ahead of several hosted enterprise-search products.[1][2]

• Custom agents that take actions. Build agents with scoped knowledge, custom instructions, and the ability to act in external applications through MCP and OpenAPI actions, from scheduled messages to end-to-end task completion.[1][2]

• Deep research and live web search. Multi-step research flows produce in-depth reports, while web search through Serper, Google PSE, Brave, SearXNG, or the built-in crawler augments internal knowledge with current public information.[1]

• Secure sandbox and artifacts. Execute code and work with intermediate files in a sandbox for complex analysis, then generate documents, graphics, and downloadable artifacts that stay attached to the conversation.[1]

• Every major model, hosted or local. Use proprietary providers or self-hosted models side by side, and keep embeddings and inference inside your own boundary when policy requires it.[1]

• Enterprise administration. Single sign-on via Google OAuth, OIDC, or SAML with SCIM provisioning, role-based access control, usage analytics, query-history auditing, PII filtering hooks, and white-labelling on Enterprise plans.[1][2]

Who will love it

• Teams whose knowledge lives in many tools. One continuously synced index replaces hunting across Slack threads, wikis, drives, and tickets, and it stays current without anyone curating uploads.[1][2]

• Security-conscious organizations. Air-gappable self-hosting keeps the document index, database, and model traffic inside infrastructure the team owns, with permission enforcement and audit trails as part of the design.[1][2]

• Operators standardizing AI across departments. An admin dashboard, shared agents, usage analytics, and a Business cloud edition at $20 per user per month give central IT something governable instead of scattered subscriptions.[1][2]

Who should avoid it

• People who want zero operations. The standard deployment is a multi-container Docker or Kubernetes stack with index, workers, inference servers, cache, and blob storage, which is genuine systems work despite the single-command installer.[1]

• Solo users wanting a free chat UI. The Lite mode covers lightweight chat, but individuals who only need multi-model chat without enterprise sync will find simpler free tools with less to run.[1][2]

• Buyers needing verified compliance marks. Permission controls and self-hosting are documented, but no SOC 2 report or certificate was verified from inspected sources, so procurement requiring one must ask the vendor directly.[1][2]

Alternatives and compatibility