LobeHub
LobeHub · LobeHub (formerly LobeChat): Self-Hostable Agent Workspace That Puts Proactive AI Teammates in Discord, Slack and Feishu
LobeHub, formerly LobeChat, is an agent workspace where you build persistent agents with their own models, memory, skills and tasks, then put them to work in the web and desktop apps or as bots in Discord, Slack, Telegram, Feishu/Lark, WeChat and other chat apps. Since v2.3.0 its bots read the channels they watch and step in on their own. It suits teams and power users who want multi-model agent teammates, either on LobeHub Cloud (free tier, paid plans from $20 a month) or self-hosted from its public code.
Best for
Teams and technical power users who want persistent, multi-model agents that join the Discord, Slack, Telegram, Feishu/Lark or WeChat chats they already use, take on tracked tasks and coding work, and can run either on a hosted plan or on their own servers
Not ideal for
People who just want a simple personal chatbot, organizations that need a strictly OSI-licensed codebase they can fork and redistribute, and teams that do not want bots joining conversations uninvited or cannot maintain a fast-moving self-hosted stack
Who it's for
Teams, communities and technical power users who want persistent multi-model agents in their team chats and a workspace for tasks, documents and coding agents, hosted or self-hosted
Capabilities
- Agents with their own system role, model, skills, MCP connectors, web search, resource library and optional structured memory that you can review, edit and delete
- Channels bind an agent to your own bot on Discord, Slack, Telegram, LINE, QQ, WeChat, Feishu or Lark, with an allowed-users list and DM, group and channel policies; one agent can serve several platforms at once, and bot tokens are encrypted before storage
- Messenger lets you reach your own agents through LobeHub's official bot on Telegram, Slack, Discord or WeChat without creating a bot, switching agents with /agents
- Proactive Bot (on by default since v2.3.0): a bot judges each message in the channels it watches and decides whether to speak up, stay quiet or wait, guided by duties, watch keywords and pace settings, with an off switch per bot
- Wake records list every message a bot judged, with the outcome, a model score and a readable reason, a 30-day summary, and buttons to mark each decision right or wrong
- Tasks assign work to an agent like an issue tracker: background runs with progress and status, comments for follow-ups, schedules or fixed intervals, labels, batch actions with undo, and optional acceptance checks with a verdict and evidence
- Agent Groups run several agents on one job in sequential, parallel, iterative or debate mode, coordinated by a built-in orchestrator
- Connect External Agents runs Claude Code, Codex, Cursor CLI, Amp, Devin CLI, Factory Droid, OpenCode, Kimi Code, OpenClaw, Hermes Agent and other runtimes on a computer linked through the desktop app or CLI, with GitHub Copilot CLI added in v2.3.2
- Cloud sandbox for running Python, JavaScript or shell commands and generating files server-side, and execution devices for working on real project files on a connected computer
- Pages for long Markdown documents written with agents, with real-time multi-person editing since v2.3.0, plus dashboards, goals and a resource library
- Models from 40+ providers, including OpenAI, Anthropic, Google, Azure OpenAI, AWS Bedrock, OpenRouter, Groq, DeepSeek, Qwen and local Ollama, LM Studio or vLLM servers, chosen per agent
- Marketplace of community agents, rated MCP servers and SKILL.md skills, with interactive MCP App views rendered inside the conversation since v2.3.2
- GitHub integration (Alpha) follows pull requests agents open: a merge accepts the work, and failing checks or reviewer feedback wake the agent to fix them; Linear issues can be imported and synced
- Web app, desktop apps for macOS, Windows and Linux, iOS and Android apps, and an `lh` command-line tool for tasks, messages, connectors and devices
Limitations
- The LobeHub Community License is based on Apache 2.0 but requires a commercial license from LobeHub to develop and distribute derivative works, and lets LobeHub change the license terms, so it is not an OSI-approved open-source license, even though a badge at the top of the README says Apache 2.0
- Since v2.3.0 every bot speaks up on its own by default; owners who want mention-only behavior have to turn off Proactive participation in each bot's wake policy
- Paid Cloud plans meter usage in 5-hour and weekly windows, model usage is billed in credits that vary by model, and every finished agent task costs an extra $0.01 from your balance
- On LobeHub Cloud, agent memory, the WeChat channel and message monitoring are not included in the Free plan
- WhatsApp and iMessage channels are listed as coming soon, and the docs say WeChat group chat support is unconfirmed
- Self-hosting needs PostgreSQL 14+ and S3-compatible storage for files, and the Vercel route needs an external database and is limited by a 10-second function timeout and no WebSocket support
- On a self-hosted deployment, the cloud sandbox needs a separate sandbox provider, and the GitHub integration needs your own GitHub App and Redis
- External agents must be installed and signed in on a computer connected through the desktop app or CLI, and the docs warn that a working directory is not a security sandbox
- Projects and the GitHub integration are Alpha features whose interface and behavior can change
- Releases are frequent (v2.3.0, v2.3.1 and v2.3.2 shipped between November 1 and 6, 2026) and run database migrations automatically on startup (0193 to 0205 across those three releases), so self-hosters need a backup and upgrade routine
Use cases
- Adding an agent to a team Discord or Slack server so it picks up unassigned requests, answers open questions and replies in the original thread
- Assigning an agent a scheduled task, such as a Monday competitor brief or a daily changelog summary, and reviewing its acceptance report before marking it done
- Running Claude Code or Codex on a connected workstation from the LobeHub chat, then letting the GitHub integration send failing CI checks back to the agent
- Putting a research agent, an analysis agent and a writing agent in one Agent Group to produce a report in sequence
- Self-hosting a private multi-model workspace for a company with Docker Compose and its own OpenAI, Anthropic or local Ollama keys
- Serving one support or community agent on Feishu, Lark, QQ or WeChat for a team working in Chinese messaging apps
Our take
LobeHub has moved well beyond the chat interface many people know as LobeChat. Its strongest idea is putting persistent agents inside the chat apps a team already uses, from Discord and Slack to Feishu, QQ and WeChat, and then backing those agents with tracked tasks, acceptance checks and a GitHub loop that turns a chat request into a reviewed pull request. Proactive Bot is the boldest choice: bots that decide for themselves when to speak are useful in a busy server, but turning it on by default means owners should write clear duties and watch the wake records in the first weeks. Two things deserve a careful look before you commit. The license is Apache-based but restricts distributing derivatives, so it is not a drop-in open-source base for a product. And the pace of change is high, with several releases a week at times, Alpha features and database migrations, which suits teams on LobeHub Cloud more easily than self-hosters without time for upgrades.
Who should use it
Teams and communities that coordinate in Discord, Slack, Telegram, Feishu/Lark or WeChat and want agents that pick up requests there, track work as tasks and hand coding jobs to Claude Code or Codex, plus technical users who want a self-hosted multi-model workspace they control.
Who should skip it
Anyone who wants a lightweight single-user chatbot, organizations that need a permissively licensed codebase to fork and redistribute, and teams that cannot tolerate bots joining conversations on their own or keep up with frequent self-hosted upgrades.
Strengths
- Agents can live in the chat apps a team already uses, including Feishu, Lark, QQ and WeChat as well as Discord, Slack and Telegram
- Proactive bots come with an audit trail: wake records explain every decision and can be marked right or wrong
- Tasks with schedules, comments and acceptance checks turn chat requests into tracked work
- Connects coding agents such as Claude Code, Codex and Copilot CLI instead of replacing them
- Wide model choice across 40+ providers, bring-your-own keys on every Cloud plan, and local models
- Both a hosted Cloud with a free tier and a documented self-hosting path, plus desktop and mobile apps
Weaknesses
- Community license restricts distributing derivative works without a commercial license
- Bots speak up unprompted by default, which needs tuning in busy channels
- Usage windows, per-model credit rates and a per-task fee make Cloud costs harder to predict
- Self-hosting is a multi-service stack (PostgreSQL, S3 storage, optional Redis and sandbox provider) with frequent migrations
- Several newer features, including Projects and the GitHub integration, are still Alpha
LobeHub pricing
Free
$0
- 500,000 credits a month
- 15 MB file storage and 100 vector entries
- 1 agent task at a time
- Bring your own provider API keys
Plus
$20
Billed monthly
- Advanced models with weekly and 5-hour usage limits
- 2 GB file storage and 2 GB cloud sandbox storage
- 2 agent tasks at a time
- Agent memory and the WeChat channel
Pro
$100
Billed monthly
- 5x the usage of Plus
- Ultra models with a separate weekly limit
- 20 GB file storage and 5 agent tasks at a time
- Priority email support
Max
$200
Billed monthly
- 10x the usage of Plus, with a 5-hour limit 20x that of Plus
- 50 GB file storage and 8 GB cloud sandbox storage
- 10 agent tasks at a time
- Priority chat and email support
Enterprise Edition
Custom
- Commercial license
- Private deployment with brand theming and user management
- Self-hosted providers and private models
- Custom integration and support
Free tier limits: The Cloud Free plan includes 500,000 credits a month, standard models, 15 MB of file storage, 3 published artifact sites and one concurrent agent task; agent memory and the WeChat channel need a paid plan.
Note: Yearly billing lowers Plus, Pro and Max to $18, $90 and $180 a month ($216, $1,080 and $2,160 a year). Model usage is measured in credits mapped to tokens, at different rates per model. A $0.01 fee is charged for each finished agent task on top of its model usage; failed tasks and tasks stopped within 30 seconds are free. Extra credits can be bought on every plan. Self-hosting the community edition has no LobeHub fee, but you pay your model providers and infrastructure.
Technical specs
Modalities
Text, Image, Voice, Files
Available models
Where LobeHub excels
An agent that picks up unassigned requests in a team chat
With Proactive Bot on and duties written, a bot bound to a Discord or Slack channel can take on a request nobody claimed and reply in the original thread, and the owner can audit each decision in wake records.
Coding work handed from chat to a connected workstation
An external agent such as Claude Code or Codex runs on a computer linked through the desktop app, and with the GitHub integration enabled, failing checks and reviewer comments on its pull request go back to that agent.
Recurring reports with a review step
A scheduled Task can produce a weekly brief, and acceptance criteria let the agent check its own work and show a verdict with evidence before you accept it.
A private workspace on company infrastructure
Docker Compose brings up LobeHub with its database, storage and search on your own server, and each agent can use your own provider keys or a local Ollama or vLLM model.
LobeHub vs. competitors
LobeHub vs. QwenPaw
QwenPaw is a free, Apache-2.0 personal assistant from the AgentScope team that you run yourself and reach through DingTalk, Feishu, WeChat, QQ, Discord, Telegram, Slack and more, with small local models available. LobeHub covers many of the same chat apps but is built as a team workspace, with proactive bots, issue-style tasks, agent groups, shared pages and a hosted Cloud, under a more restrictive community license.
LobeHub vs. OpenClaw
OpenClaw is a free, open-source personal assistant that runs locally and works through 20+ messaging apps, including WhatsApp, Signal and iMessage, which LobeHub lists only as coming soon or does not list. LobeHub is a multi-user workspace with tasks, agent groups and a hosted plan, and it can connect an existing OpenClaw install as one of its external agents.
LobeHub vs. Hermes Agent
Hermes Agent from Nous Research is an open-source agent that runs on your own server, builds layered memory as Markdown files and creates skills from experience, reached through 20+ messaging platforms. LobeHub focuses on shared team workflows rather than one persistent personal agent, and it can add Hermes as an external agent inside its workspace.
LobeHub vs. Town
Town gives each person at work a hosted assistant connected to email, calendar, Slack and 50+ tools that asks before it sends anything, as closed SaaS. LobeHub is broader and more technical: agents bound to bots in many chat apps, coding agents on connected computers, a choice of 40+ model providers, and a self-hosting option.
Frequently asked questions
Is LobeHub the same as LobeChat?
Yes. LobeHub is the current name of the project formerly called LobeChat, and the GitHub repository is now lobehub/lobehub. It has grown from a multi-model chat interface into an agent workspace with persistent agents, tasks, agent groups, pages and bots that join team chat apps; the LICENSE file still refers to the product as LobeChat.
Is LobeHub open source?
Its source code is public, but under the LobeHub Community License rather than a standard open-source license. The license is based on Apache 2.0 and allows commercial use, including running LobeHub as a frontend and backend service without modifying the code. Developing and distributing a derivative work requires a commercial license from LobeHub, and LobeHub may make the terms stricter or more relaxed. That makes it source-available rather than OSI open source. A badge at the top of the GitHub README says Apache 2.0, but the LICENSE file and the README's own license section name the LobeHub Community License.
How much does LobeHub cost?
Self-hosting is free under the community license, and you pay your model providers. LobeHub Cloud has a Free plan with 500,000 credits a month, 15 MB of file storage and one concurrent agent task. Plus is $20 a month, Pro $100 and Max $200 ($18, $90 and $180 a month billed yearly), with usage measured in 5-hour and weekly windows, more storage and more concurrent tasks. Every plan adds a $0.01 fee per finished agent task, extra credits can be bought, and an Enterprise Edition with a commercial license and private deployment is sold through sales.
Can LobeHub be self-hosted?
Yes. The recommended route is Docker Compose, which starts LobeHub with PostgreSQL, Redis, RustFS file storage, SearXNG search and the agent gateway on Linux, macOS or Windows through WSL 2. One-click deploys are offered for Vercel, Zeabur, Sealos, RepoCloud and Alibaba Cloud; Vercel needs an external PostgreSQL database and S3-compatible storage and has a 10-second function timeout. You need at least one AI provider API key, and features such as the cloud sandbox and the GitHub integration need extra services.
Which chat apps can LobeHub agents join?
With Channels, an agent can run as your own bot on Discord, Slack, Telegram, LINE, QQ, WeChat, Feishu and Lark. Group chats work on all of them except WeChat, where the docs say group support is unconfirmed, and WeChat needs a paid plan on LobeHub Cloud. WhatsApp and iMessage are listed as coming soon. Alternatively, Messenger lets you talk to your agents through LobeHub's own bot on Telegram, Slack, Discord or WeChat.
Will a LobeHub bot reply without being @-mentioned?
By default, yes. Since v2.3.0, Proactive Bot is on for every bot: it judges each message in the channels it watches and decides whether to speak up, stay quiet or wait, for example when a message falls within the duties its owner wrote or someone leaves a question unanswered. The owner can turn off Proactive participation in the channel settings' wake policy, after which only watch keywords wake it. The Wake records tab shows each decision with a reason.
Can LobeHub run Claude Code or Codex?
Yes, as external agents. You install and sign in to the agent CLI on a computer, connect that computer through the LobeHub desktop app or CLI, and add the agent through Connect External Agents. The catalog includes Claude Code, Codex, Cursor CLI, Amp, Devin CLI, Factory Droid, OpenCode, Kimi Code, Qoder and others, plus OpenClaw and Hermes Agent, and v2.3.2 added GitHub Copilot CLI. The agent keeps its own permission system, and LobeHub shows its replies, tool calls and file changes in the chat.
Integrations & fit
Alternatives to consider
AgentScope
Technical users and small teams who want a self-hosted assistant in DingTalk, Feishu, WeCom, WeChat, or QQ, offline use with small local models, and built-in tool guards with an OS-level shell sandbox
Free
OpenClaw Foundation
Technical users who want a private, self-hosted AI assistant they fully control, running through their own messaging apps and model of choice
Free
Nous Research
Developers and technical power users who want a persistent, self-hosted AI agent with compounding memory and full control over their infrastructure and model choice
FreeAbout LobeHub
LobeHub started as LobeChat, a popular self-hosted multi-model chat interface, and is now positioned as a platform for "super-agents on your team". The core unit is an agent with a system role, a chosen model, skills, MCP connectors, web search, a resource library and, when enabled, structured memory you can review, edit and delete. Agents can take on Tasks, which work like Linear or GitHub issues with an agent as the assignee: a task runs in the background, reports progress, can run on a schedule, takes follow-ups in comments, and can be checked against acceptance criteria before delivery. Agent Groups put several agents on one job in sequential, parallel, iterative or debate mode under an orchestrator, and Pages let people and agents write long Markdown documents together, with real-time co-editing since v2.3.0. The distinctive part is where the agents live. Through Channels, each agent can be bound to your own bot on Discord, Slack, Telegram, LINE, QQ, WeChat, Feishu or Lark, with per-user and per-channel access rules; through Messenger, you can instead talk to your agents via LobeHub's official bot on Telegram, Slack, Discord or WeChat. Since v2.3.0 (November 1, 2026) Proactive Bot is on by default: a bot judges each message in the channels it watches and decides whether to speak up, and a Wake records tab shows every decision with a reason so the owner can mark it right or wrong. LobeHub can also connect agent runtimes rather than just models: Claude Code, Codex, Cursor, Amp, Devin, OpenCode, OpenClaw, Hermes and other CLIs installed on a computer linked through the desktop app or the `lh` CLI, with GitHub Copilot CLI added in v2.3.2. Models come from 40+ providers, including OpenAI, Anthropic, Google, AWS Bedrock, OpenRouter, DeepSeek and local Ollama, LM Studio or vLLM servers. LobeHub Cloud runs on credits, with a free tier and Plus, Pro and Max plans at $20, $100 and $200 a month, plus a $0.01 fee per agent task. Self-hosting uses Docker Compose (bundling PostgreSQL, Redis, S3-compatible storage and search) or Vercel and similar platforms. The tradeoffs: the code is public under the LobeHub Community License, which is based on Apache 2.0 but requires a commercial license to distribute derivative works, so it is not open source in the OSI sense. Bots that speak up unprompted need tuning, several features (Projects, the GitHub integration) are still Alpha, and frequent releases with database migrations put the burden of upgrades on self-hosters.
Updates from LobeHub
LobeHub v2.3.2 (November 6, 2026) adds GitHub Copilot CLI as an external agent alongside Claude Code and Codex, renders interactive MCP App views inside the conversation, and adds task labels, batch task actions with undo and a recycle bin for deleted tasks. Conversations now open as channels, and managed cloud devices can be connected with `lh connect --managed`.
LobeHub v2.3.1 (November 3, 2026) lets you switch to auto-approve from an approval card, fixes duplicate tool execution on resume, adds Linear comment sync and completed-issue imports, and adds paged wake records and nightly bot retrospectives.
LobeHub v2.3.0 (November 1, 2026) makes every bot read the channels it watches and decide for itself when to speak up, with a per-bot off switch and a Wake records tab that explains each decision. The release also brings real-time page collaboration, dashboards, workspace-level goals and a persistent sandbox that is out of preview.
Are you the founder? Claim this listing →