Onyx

Onyx is an open-source enterprise AI platform for connected workplace knowledge. It indexes connected content for document search and AI chat grounded in organizational knowledge, with adjacent capabilities for agents, tools, web search, code execution, and API or MCP access. Cloud and self-hosted deployments are available, including air-gapped self-hosted installations.

Deployment choice matters: Onyx Standard includes vector and keyword RAG indexing plus connector synchronization, while Onyx Lite cannot index knowledge from connected applications.

Onyx separates discovery from answer generation. Search returns ranked source documents as source cards; Chat generates cited answers. Retrieval features include hybrid keyword and vector search, advanced RAG, contextual retrieval, and query expansion.

For operational traceability, Query History can show messages, user feedback, and reference documents used for an answer, subject to plan and visibility settings. Citations and source visibility help with verification, but they do not guarantee factual correctness or hallucination-free answers.

Onyx supports connectors for knowledge bases, cloud storage, ticketing, messaging, meetings, sales, and code repositories, plus file upload and an ingestion API. The pricing page lists 40+ app connectors.

Source permission synchronization is available for named connectors on Business, Enterprise, and Cloud plans. Behavior varies by connector: Salesforce does not synchronize document permissions, and source scope and exclusions also vary. Content refresh runs every 30 minutes by default, deleted-content pruning defaults to seven days, and permission changes can take from a few minutes to an hour to appear.

Administrators can manage models, providers, connectors, indexing, search features, spending limits, and model access controls. Administrator and group-scoped roles let Group Managers manage members and group-shared resources without organization-wide administrative powers.

Usage dashboards track active users, spending, tokens, and model and feature breakdowns. Query History supports filtering, visibility modes, reference document inspection, and CSV export. Self-hosted deployments emit structured audit events that can be forwarded to SIEM platforms, including Splunk, Microsoft Sentinel, Elastic, Google Chronicle, and AWS Security Lake. Enterprise adds SCIM and group sync, custom roles and permissions, configurable limits, custom deployments, region-specific processing, priority support, and enterprise SLAs.

Onyx fits security-conscious organizations and technical teams whose knowledge is spread across systems like Drive, Slack, Confluence, SharePoint, GitHub, Jira, CRM, support tools, and document repositories. Its self-hosting, controlled model choices, APIs, custom connectors, and configuration options are particularly relevant where deployment control matters.

Buyers should test connector-specific permissions and exclusions, expected content freshness, and the edition required for their use case. Organizations that need every connector to mirror source ACLs, universally real-time retrieval, or indexed enterprise search on Lite should validate those requirements before rollout.