Home

ChatHub

Self-hosted AI chat for Docker and PostgreSQL — a fork of LobeChat that grew into its own product. Current GA is 2.0.

v1.0.0 was a rebranded, self-hosted LobeChat (credentials login, Tools Hub, MCP OAuth, per-provider gear). 2.0 is the same Docker upgrade path, but the runtime contract changed: leave is not Stop, Knowledge Base has its own embedding keys, Python and MarkItDown are optional sidecars, and there is no Electron or browser-local database.

Product story and deploy example: repository README. Full 1.0.0→2.0.0 write-up: v2.0.0 release. This page is the procedure index.

	Upstream LobeChat	ChatHub 2.0
Deploy	Vercel / Docker / Desktop	Docker + PostgreSQL only
Chat after you leave	Cancels with the tab	Server worker continues
Knowledge embeddings	Implicit chat-provider defaults	Dedicated RAG provider
Sandbox	Not a first-class sidecar	Optional OpenSandbox sibling
Client	Includes desktop / local DB editions	Browser and PWA against the server

---

Features

Core (inherited from LobeChat)

• Multi-model chat — OpenAI, Anthropic, Google, Moonshot/Kimi, MiniMax, DeepSeek, Xiaomi MiMo, xAI Grok, OpenAI-compatible (custom URL + API key + models), Ollama
• Knowledge base with RAG (file upload, chunking, vector search)
• MCP plugin system with one-click installation
• Multi-user management with NextAuth / OIDC / Clerk
• Image generation (DALL-E, GPT-Image, Gemini Image)
• Chain-of-thought, branching conversations, artifacts support
• TTS / STT voice conversation
• Real-time search integration

ChatHub Additions

Chat

• Built-in credentials login — username/password or token login page, no external OAuth service needed. Authentication
• Background conversation generation — replies continue after you close the tab; Stop and destructive history actions cancel matching work. Details
• Pasted text cards — large dumps collapse to a compact PASTED chip in the composer and a matching card in the topic; the model still receives the full text. Details
• Context Export — capture the context prepared for the next model request. Details
• Memory and context compaction — assistant presets, auto-compact, manual compact, daily note, archives, assistant memory. Details
• Chat Instruction — standing user instruction applied across chats for that account
• Input Markdown rendering — enable or disable live composer formatting in Settings → Common → Chat Appearance

Knowledge

• Knowledge Base and RAG — external OpenAI/Cohere/Voyage embeddings, explicit readiness, pgvector indexing, and supported-document filtering. Setup and behavior
• MarkItDown sidecar — optional companion that converts uploads to structured Markdown before chunking. Details

Tools

• Tools Hub — dedicated sidebar section for Picbed, API Tester, Password Generator. Details
• Picbed — image and video hosting with S3, select/drag/paste upload, auto URL copy
• API Tester — full REST API client
• Sandbox — shell, git, Node.js, and Python in one container per conversation, on OpenSandbox. Setup
• MCP OAuth 2.1 — auto-discovery, popup flow, server-side token injection. Details
• MCP library — Settings page for installed HTTP MCP plugins. Details
• Agent Skills — globally installed SKILL.md bundles with lazy load_skill. Details

Media

• Artifacts gallery — account-scoped gallery of generated images at /artifacts. Details
• Inline SVG diagrams — themed SVG artifacts plus rendered HTML / SVG / Mermaid code blocks. Details

Providers

• Extended model bank — latest Claude, GPT-5.x, Gemini 3.x, Kimi K2.x/K3, MiniMax, DeepSeek, Zhipu GLM, Xiaomi MiMo, xAI Grok
• Model extension options — per-provider gear panels (Moonshot, MiniMax, Anthropic, OpenAI, DeepSeek, Xiaomi MiMo, xAI). Details
• OpenAI-compatible prompt cache — provider-settings matrix. Details
• ChatGPT / Codex subscription — per-user Sign in with ChatGPT on Settings → OpenAI. Details
• xAI / Grok — Console API key plus Settings-only SuperGrok device login and weekly meters. Details

Ops

• JSON backup and restore — versioned account export/import under Settings → Storage. Details
• PWA — installable app with separate any and maskable icons
• Debug switches — privacy-safe diagnostics. Details