Onyx: Open-source enterprise chat UI for any LLM with RAG, tools, and deep research

Read Articleadded Nov 25, 2025

Onyx is an open-source, model-agnostic chat UI that brings enterprise features and integrated tools like RAG, web search, MCP, deep research, and a secure code interpreter. Born from Danswer, it focuses on consumer-grade UX for everyday work while supporting complex agentic workflows and on-prem, air-gapped deployments. The team highlights critical engineering tactics for context management and model-specific tool behavior, and reports early traction with large enterprises.

Key Points

  • Pivot from Danswer: Users primarily wanted a secure, high-quality chat UI over full enterprise search, prompting the creation of Onyx.
  • Focus on UX: Matching and exceeding consumer-grade chat experiences for work use is hard but essential for adoption.
  • Full toolchain integration: RAG, web search, MCP, deep research, memory, assistants, and a secure, model-agnostic code interpreter.
  • Enterprise-ready: RBAC, SSO, permission syncing, and easy on-prem/air-gapped hosting for sensitive industries.
  • Engineering learnings: Use end-of-message “Reminder” prompts for instruction adherence and adapt tool behavior to model tendencies.

Sentiment

The overall sentiment is cautiously positive but mixed with significant skepticism and critical feedback. While there's an acknowledgment of Onyx's value proposition for enterprise AI chat and secure deployments, particularly its strong RAG capabilities, there are notable concerns regarding its 'open core' licensing model, product maturity, and differentiation in a crowded market. The founders' active engagement in the discussion helped clarify some points.

In Agreement

  • Many enterprises struggle and often fail to build effective internal AI chat solutions because product managers misunderstand user needs for simple chat interfaces over complex agent creation, validating Onyx's focused approach.
  • The value proposition for Onyx, particularly its robust RAG and connector suite, is highly appreciated, stemming from the creators' enterprise search background.
  • Onyx fulfills a crucial need for secure, on-premise, or air-gapped LLM deployments, especially in highly regulated industries or regions where major cloud providers face limitations.
  • The open-source nature of Onyx's core components is seen as a positive, allowing for white-labeling and community contributions to connectors.
  • The ability to compete with large players on pricing, especially for customers with large staff and low usage, by offering BYOK deployments, is a key advantage for Onyx.

Opposed

  • Significant confusion and criticism exist regarding Onyx's "open core" licensing model (MIT for core, proprietary for enterprise extensions), with some arguing it's "source-available" rather than truly "open source" by OSI definitions.
  • Concerns were raised about the overall maturity and "unbaked" feel of some features, particularly regarding document tracking and mapping within the UI.
  • Some commenters questioned the necessity of self-hosting for security, pointing to the SOC 2 compliance of major cloud LLM providers.
  • The product's differentiation from numerous existing AI chat UI projects is questioned, with some suggesting a lack of a clear competitive moat.
  • The choice of "Onyx" as a product name is criticized for its similarity to "ONNX," an already established term in the ML/AI community.
  • Prosumer users criticized the simplified UI, arguing it sacrifices the granular control and advanced settings offered by other local model interfaces like oobabooga/sillytavern.