Your Chief Agent Operator — build, schedule, and collaborate with an entire AI team in one self-hostable workspace.
LobeHub (formerly LobeChat) is an open-source AI workspace that treats agents as the fundamental unit of work. Instead of a simple chat interface, LobeHub gives you a full operational environment where you hire agents, assign them skills, schedule them to run autonomously, and organize their output across projects and pages — all from a single platform you can self-host.
The platform connects to over 80 LLM providers — Anthropic, OpenAI, Google Gemini, Deepseek, Bedrock, Azure, Hugging Face, and dozens more — through a unified model runtime. Agents can be equipped with a library of over 10,000 MCP-compatible tools and plugins covering web browsing, file access, code execution, memory management, and external API integrations.
LobeHub introduces multi-agent collaboration through Agent Groups, where the system assembles the right agents for a task, runs them in parallel, and iterates on results in a shared context. Personal Memory gives agents a structured, editable understanding of the user’s preferences and working style, enabling continual adaptation over time rather than one-off responses.
Built on Next.js with a PostgreSQL backend via Drizzle ORM, LobeHub can be deployed to Vercel, Docker, Zeabur, Sealos, or Alibaba Cloud. The project has amassed over 78,000 stars and ships multiple canary releases per day, reflecting an extraordinarily active development pace.
Architecture LobeHub is structured as a Next.js monorepo using pnpm workspaces, separating the main application shell from a deep tree of independently versioned packages — agent runtimes, model providers, built-in tools, chat adapters, and database layers each live in their own package boundary. The application layer uses file-based routing from the Next.js App Router with a clear split between server components, API routes handled via tRPC and Hono, and a rich client-side React store architecture built on Zustand. Agents, tools, and skills are decoupled through runtime registries rather than hard-coded imports, which lets the platform compose arbitrary combinations of models, tools, and memory backends without rewriting core logic.
Tech Stack The frontend is built with Next.js 15 and React, styled with Ant Design Pro components augmented by the project’s own @lobehub/ui design system and Emotion CSS-in-JS. State management relies on Zustand stores organized around domain slices — agent, chat, device, global, and more. The server layer runs on Node.js with tRPC for type-safe RPC calls and Hono for edge-compatible HTTP endpoints. Persistence is PostgreSQL accessed through Drizzle ORM with schema-driven migrations. Auth is handled by better-auth with passkey and Expo support. The model integration layer abstracts over 80 providers through a unified provider registry in @lobechat/model-runtime, with individual provider adapters for Anthropic, OpenAI, Google, Bedrock, Azure, Deepseek, Groq, Hugging Face, Cloudflare, and many more. Tests run via Vitest with extensive mocks and E2E coverage.
Code Quality With over 1,700 test files spanning unit, integration, and mock layers, LobeHub demonstrates a serious commitment to testing at scale. The TypeScript codebase is strictly typed throughout, with ESLint, Prettier, and Stylelint enforced in CI. The monorepo uses Knip for dead-code elimination and Codecov for coverage tracking. Commit discipline is enforced via commitlint, and the project ships nightly canary builds from active PRs in addition to weekly stable releases. Inline documentation is moderate — the code is readable and consistently structured — but the sheer scale of the codebase means some of the newer packages are lighter on comments than the mature core.
What Makes It Unique LobeHub’s genuine innovation is treating agents as persistent, schedulable teammates rather than stateless API call wrappers. The combination of Agent Groups (parallel multi-agent collaboration with shared context), the editable white-box Personal Memory architecture, the IM Gateway (bringing agents into existing messaging platforms), and the 10,000+ MCP tool library creates an operational layer that doesn’t exist in most AI chat applications. The Chief Agent Operator framing — where you define an AI team’s structure and it executes asynchronously — is a meaningful architectural departure from the prompt-response paradigm, enabling workflows that continue without the user being present.
LobeHub is released under the LobeHub Community License, which is based on Apache 2.0 with one significant addition: if you want to build and distribute a derivative commercial product based on LobeHub’s source code, you need to obtain a commercial license from the LobeHub team. Running LobeHub as-is — including as a frontend and backend service for your own organization — is permitted commercially without a separate license. This means teams can self-host freely for internal use, but ISVs or SaaS builders who want to ship a modified version to paying customers need to contact hello@lobehub.com. The practical implication for most developers is that self-hosting for personal or organizational use carries no licensing friction, but forking to create a competing product does.
Operating LobeHub yourself means running a Next.js application backed by a PostgreSQL database. Docker Compose is the most straightforward path and is well-documented. You are responsible for database backups, SSL termination, environment variable management (API keys for every LLM provider you want to support, auth secrets, S3 credentials if using file uploads), and scaling the application layer as load grows. The platform’s heavy feature set — multi-agent orchestration, real-time streaming, MCP tool execution, file loaders, knowledge bases — means the operational surface is meaningfully larger than a simple chatbot. Canary releases drop multiple times per day, so you will want to pin to stable releases rather than following HEAD if you value predictability.
Compared to using lobehub.com’s hosted version, self-hosting means you handle uptime, upgrades, and support yourself. The hosted cloud tier adds managed infrastructure, guaranteed availability, and likely priority access to new features before they stabilize in the community release. For organizations evaluating the trade-off: self-hosting gives you full data sovereignty and removes all API key intermediation — your LLM calls go directly from your server to the provider — which is the primary reason serious privacy-conscious teams choose this route over the cloud offering.
Automation · Productivity · AI Assistants
Build, deploy, and run autonomous AI agents that automate complex multi-step workflows using a visual block-based graph editor.
No Code Platforms · AI Development · Developer Tools
Visual LLM workflow platform with RAG pipelines, agent capabilities, and model management for building production AI applications.
Productivity · Project Management · Collaboration
The open-source AI workspace that puts your data, your rules — with local LLMs, CRDT collaboration, and full self-hosting built in.