Onyx — Self-Hosted AI Chat with 40+ Connectors

What it is

Onyx (formerly Danswer) is a self-hosted AI chat platform that connects to your organization's internal knowledge. It supports 40+ connectors for tools like Slack, Confluence, Google Drive, Notion, GitHub, Jira, and more. Onyx uses retrieval-augmented generation (RAG) to ground AI answers in your actual documents, reducing hallucination.

Onyx is built for teams that need enterprise search and AI chat with full data control. If you want ChatGPT-like functionality over your internal documents without sending data to third-party services, Onyx keeps everything on your infrastructure.

How it saves time or tokens

Onyx replaces the pattern of manually searching across multiple tools (Slack, Confluence, Google Drive) to find information. You ask a question in natural language and Onyx retrieves relevant documents from all connected sources, then generates an answer with citations. The RAG pipeline ensures answers reference actual internal documents rather than general knowledge, reducing follow-up clarification. Custom agents let you create specialized assistants for different teams or workflows.

How to use

1. Install Onyx: curl -fsSL https://onyx.app/install_onyx.sh | bash.
2. Open http://localhost:3000 and configure your LLM provider (OpenAI, Anthropic, or local models).
3. Connect knowledge sources through the admin panel and start asking questions.

Example

# Quick install with Docker
curl -fsSL https://onyx.app/install_onyx.sh | bash

# Or manual Docker Compose setup
git clone https://github.com/onyx-dot-app/onyx.git
cd onyx/deployment/docker_compose
docker compose -f docker-compose.dev.yml up -d

# Access the UI at http://localhost:3000
# 1. Set up your LLM provider (API key)
# 2. Add connectors (Slack, Confluence, etc.)
# 3. Ask questions grounded in your documents

Related on TokRepo

• RAG tools -- retrieval-augmented generation tools and frameworks
• Self-hosted tools -- tools you can run on your own infrastructure

Common pitfalls

• Not configuring connector sync schedules. By default connectors may sync infrequently. Set appropriate intervals so the AI has access to recent documents.
• Using a weak LLM for complex queries. Onyx supports multiple LLM providers. For nuanced enterprise questions, use a capable model rather than a small local model.
• Ignoring access controls. Onyx supports document-level permissions from connected sources. Configure these to prevent users from seeing documents they should not have access to.