🔥 onyx-dot-app/onyx
Open Source AI Platform - AI Chat with advanced features that works with every LLM — Trending on GitHub today with 582 new stars.
Open Source AI Platform
Onyx is a feature-rich, self-hostable Chat UI that works with any LLM. It is easy to deploy and can run in a completely airgapped environment.
Onyx comes loaded with advanced features like Agents, Web Search, RAG, MCP, Deep Research, Connectors to 40+ knowledge sources, and more.
Tip
Run Onyx with one command (or see deployment section below):
curl -fsSL https://onyx.app/install_onyx.sh | bash
-
🤖 Custom Agents: Build AI Agents with unique instructions, knowledge and actions.
-
🌍 Web Search: Browse the web with Google PSE, Exa, and Serper as well as an in-house scraper or Firecrawl.
-
🔍 RAG: Best in class hybrid-search + knowledge graph for uploaded files and ingested documents from connectors.
-
🔄 Connectors: Pull knowledge, metadata, and access information from over 40 applications.
-
🔬 Deep Research: Get in depth answers with an agentic multi-step search.
-
▶️ Actions & MCP: Give AI Agents the ability to interact with external systems.
-
💻 Code Interpreter: Execute code to analyze data, render graphs and create files.
-
🎨 Image Generation: Generate images based on user prompts.
-
👥 Collaboration: Chat sharing, feedback gathering, user management, usage analytics, and more.
Onyx works with all LLMs (like OpenAI, Anthropic, Gemini, etc.) and self-hosted LLMs (like Ollama, vLLM, etc.)
To learn more about the features, check out our documentation!
🚀 Deployment
Onyx supports deployments in Docker, Kubernetes, Terraform, along with guides for major cloud providers.
See guides below:
-
Docker or Quickstart (best for most users)
-
Kubernetes (best for large teams)
-
Terraform (best for teams already using Terraform)
-
Cloud specific guides (best if specifically using AWS EKS, Azure VMs, etc.)
Tip
To try Onyx for free without deploying, check out Onyx Cloud.
🔍 Other Notable Benefits
Onyx is built for teams of all sizes, from individual users to the largest global enterprises.
-
Enterprise Search: far more than simple RAG, Onyx has custom indexing and retrieval that remains performant and accurate for scales of up to tens of millions of documents.
-
Security: SSO (OIDC/SAML/OAuth2), RBAC, encryption of credentials, etc.
-
Management UI: different user roles such as basic, curator, and admin.
-
Document Permissioning: mirrors user access from external apps for RAG use cases.
🚧 Roadmap
To see ongoing and upcoming projects, check out our roadmap!
📚 Licensing
There are two editions of Onyx:
-
Onyx Community Edition (CE) is available freely under the MIT license.
-
Onyx Enterprise Edition (EE) includes extra features that are primarily useful for larger organizations. For feature details, check out our website.
👪 Community
Join our open source community on Discord!
💡 Contributing
Looking to contribute? Please check out the Contribution Guide for more details.
Sign in to highlight and annotate this article

Conversation starters
Daily AI Digest
Get the top 5 AI stories delivered to your inbox every morning.
More about
githubtrendingopen-source
I Traced a "Cute" Minecraft Phishing Site to a C2 Server in Chicago
Hello community! As an IT engineering student, I recently conducted a technical investigation into an active threat targeting the gaming community (specifically Minecraft players). What appeared to be a harmless "cute" website turned out to be a Phishing and Malware-as-a-Service (MaaS) infrastructure. Here is a technical breakdown of my findings: PHISHING AND MALWARE SPREAD THROUGH DISCORD The primary domain identified is owocraft.com. At first glance, it uses Tailwind CSS and a Turkish-coded template (identified by source code comments such as /* Sayfa Fade-in Animasyonu */). The main deception is a download button for a fake "Launcher" that actually points to a malicious .rar file hosted on Dropbox (ID: 3d1d505ajob480fkdnpm3). This file contains a Discord Token Stealer. Unmasking the Inf

NevaMind AI: Advanced Memory for Proactive Agents
Unveiling memU: A Sophisticated Memory Solution for 24/7 Proactive AI Agents NevaMind AI is thrilled to introduce memU , an open-source project dedicated to providing advanced memory functionalities for AI agents operating around the clock. Designed with the demands of proactive systems in mind, such as the principles behind Moltbot (ClawDBot), memU aims to be a cornerstone for developing more intelligent and responsive AI. Why memU? In the rapidly evolving landscape of Artificial Intelligence, robust memory management is paramount for agents that need to perform complex tasks, maintain context over long interactions, and learn continuously. memU addresses this critical need by offering: 24/7 Proactive Operation : Ensures agents are always ready, minimizing latency and maximizing efficienc

Stop leaking your .env to AI! I built a Rust/Tauri Secret Manager to inject API keys safely 🛡️
The AI Editor Problem 🤖 We all love using AI assistants like Cursor, GitHub Copilot, and Claude. But let's be honest: they brought a terrifying new security risk to our local development workflow. Have you ever worried that your AI assistant might accidentally read your local .env file and send your raw database passwords or OpenAI API keys to the cloud? Or maybe you were screen-sharing and accidentally opened your .env file for everyone to see? To solve this problem once and for all, I built Kimu — an open-source, hybrid CLI GUI secret manager powered by Rust and Tauri. 👉 Check out Kimu on GitHub! 🗝️ The Magic of Kimu: Use Placeholders, Not Passwords With Kimu, you no longer need to write actual sensitive information in your .env files. Instead, you use simple placeholders. ❌ Before (D
Knowledge Map
Connected Articles — Knowledge Graph
This article is connected to other articles through shared AI topics and tags.
More in Open Source AI

NevaMind AI: Advanced Memory for Proactive Agents
Unveiling memU: A Sophisticated Memory Solution for 24/7 Proactive AI Agents NevaMind AI is thrilled to introduce memU , an open-source project dedicated to providing advanced memory functionalities for AI agents operating around the clock. Designed with the demands of proactive systems in mind, such as the principles behind Moltbot (ClawDBot), memU aims to be a cornerstone for developing more intelligent and responsive AI. Why memU? In the rapidly evolving landscape of Artificial Intelligence, robust memory management is paramount for agents that need to perform complex tasks, maintain context over long interactions, and learn continuously. memU addresses this critical need by offering: 24/7 Proactive Operation : Ensures agents are always ready, minimizing latency and maximizing efficienc

Gemma 4 for 16 GB VRAM
I think the 26B A4B MoE model is superior for 16 GB. I tested many quantizations, but if you want to keep the vision, I think the best one currently is: https://huggingface.co/unsloth/gemma-4-26B-A4B-it-GGUF/blob/main/gemma-4-26B-A4B-it-UD-IQ4_XS.gguf (I tested bartowski variants too, but unsloth has better reasoning for the size) But you need some parameter tweaking for the best performance, especially for coding: --temp 0.3 --top-p 0.9 --min-p 0.1 --top-k 20 Keeping the temp and top-k low and min-p a little high, it performs very well. So far no issues and it performs very close to the aistudio hosted model . For vision use the mmproj-F16.gguf. FP32 gives no benefit at all, and very importantly: --image-min-tokens 300 --image-max-tokens 1024 Use a minimum of 300 tokens for images, it inc

Untitled
You have 50 models. Each trained on different data, different domain, different patient population. You want them to get smarter from each other. So you do the obvious thing — you set up a central aggregator. Round 1: gradients in, averaged weights out. Works fine at N=5. At N=20 you notice the coordinator is sweating. At N=50, round latency has tripled, your smallest sites are timing out, and your bandwidth budget is gone. You tune the hell out of it. Same ceiling. This is not a configuration problem. This is an architecture ceiling. The math underneath it guarantees you hit a wall. A different architecture changes the math. The combinatorics you are not harvesting Start with a fact that has nothing to do with any particular framework: N agents have exactly N(N-1)/2 unique pairwise relati

Discussion
Sign in to join the discussion
No comments yet — be the first to share your thoughts!