LobeChat
The polished, self-hostable multi-model chat client — every provider, one beautiful UI.
About
LobeChat (by LobeHub) is an open-source-style chat application that puts GPT, Claude, Gemini, DeepSeek, local models and 40+ other providers behind a single refined interface. It runs as a hosted app (app.lobehub.com) or self-hosted via one-click Vercel/Zeabur/Sealos/Alibaba deploys or Docker, with your own API keys. Beyond chat, it ships a knowledge base for RAG over your documents, a plugin and MCP-compatible tool ecosystem, an agent builder with custom system prompts and tools, multi-session/topic organization, TTS/STT voice, and art/image generation. Note the licensing nuance: since v1.0 it is distributed under the custom LobeHub Community License (Apache-2.0-based with a commercial-derivative restriction), which is source-available rather than OSI open source. The project is currently evolving from a chat client into the broader LobeHub 'agent operator' platform.
What it can do
- **Multi-provider chat**: OpenAI, Claude, Gemini, DeepSeek, Qwen, GLM and 40+ providers in one UI, switchable per session
- **Agent builder**: custom assistants with system prompts, tools and knowledge; shareable agent configs
- **Knowledge base / RAG**: upload documents, chat with file, per-agent file search
- **Plugin + MCP ecosystem**: tools and MCP-compatible plugins extend what agents can do
- **Sessions & topics**: multi-session management, topic branching, history search
- **Voice**: TTS/STT for spoken conversations
- **Art generation**: image generation via configured providers
- **Self-hosting**: Docker Compose, Vercel/Zeabur/Sealos/Alibaba one-click, env-var driven config, desktop app
Setup
Option A — Vercel one-click (no server)
- Fork/star the repo, click the Vercel deploy button in the README
- Set the
OPENAI_API_KEYenvironment variable (or other provider keys) during deploy - Open your deployment URL and start chatting
Option B — Docker (recommended for full self-host)
mkdir lobehub-db && cd lobehub-dbbash <(curl -fsSL https://lobe.li/setup.sh)— initializes the database/infrastructuredocker compose up -d— starts the LobeHub service (chat + server DB)- Open http://localhost:3210 and add your provider API keys in Settings
Option C — npm / source
npm install -g @lobehub/chator clone the repo and runbun install && bun run dev- Configure provider keys via environment variables or the in-app settings panel
Detailed env-var reference (custom models, proxies, auth): https://lobehub.com/docs/self-hosting/server-database/docker
Prerequisites
- Your own LLM API key(s): OpenAI, Anthropic, Google, DeepSeek, etc. (BYOK — you pay the provider) - For Docker: Docker + docker-compose and a machine with a few GB RAM - For Vercel: a Vercel account and GitHub login - Local models work through any OpenAI-compatible endpoint (e.g. Ollama at http://localhost:11434)
Use cases
Works with
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: 83,175 GitHub stars (log-scaled vs catalog peers); blended with ~122,754 monthly package downloads; last push 0d ago; 100 releases/12mo; 1058 open issues. Manual review: repo renamed lobehub/lobe-chat → lobehub/lobehub (rebranded LobeHub); '10,000+ skills/plugins' is README claim. License: LobeHub Community License — custom, NOT OSI (commercial distribution of derivatives needs permission).
★ 83,175 GitHub stars
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 8.6/10
- Maintenance · recency, releases, issue load
- 9.5/10
- Openness · license + docs
- 6.5/10
- Manual review · sanity read + red-flag check
- 8.0/10
- · 83,175 GitHub stars (log-scaled vs catalog peers)
- · blended with ~122,754 monthly package downloads
- · last push 0d ago; 100 releases/12mo; 1058 open issues
- · manual review: repo renamed lobehub/lobe-chat → lobehub/lobehub (rebranded LobeHub); '10,000+ skills/plugins' is README claim
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.