dalao.ai

LobeHub

A workspace for organizing and running teams of AI agents

About LobeHub

LobeChat is now LobeHub. Once a self-hostable AI chat UI, it was renamed in v2.0.0 (2026-01-27) and recast from a chat framework as a workspace for managing teams of AI agents; the repo moved from lobehub/lobe-chat to lobehub/lobehub. As of 2026-10-25: 83,471 stars, latest stable v2.2.20 (2026-10-18). Hosted LobeHub Cloud is a credit subscription (free tier: 500,000 credits a month); the community edition self-hosts at no software cost. The license is the LobeHub Community License, not plain Apache 2.0: distributing a derivative work requires buying a commercial license first.

EditorialUpdated 10 min read

What it is

LobeHub is developed and maintained by LobeHub LLC, the copyright holder named in its LICENSE. It was open-sourced in 2023 as a good-looking multi-model chat web app. The 1.0 release (2024-06-19) added a server-side Postgres database, multi-user auth and a beta of the paid cloud. The 2.0 release (2026-01-27) brought the rename plus multi-agent collaboration, Agent Builder, Pages, memory and a rebuilt knowledge base.

The current README pitches it as a "Chief Agent Operator". In its own words, it hires, schedules and reports on a team of agents running 7×24 (vendor claim). The listed features are:

  • Agent Builder: turns a one-line description into an agent.
  • Skills marketplace: "10,000+ Skills" (vendor claim) plus MCP plugins.
  • Agent Groups: several agents work in one conversation.
  • Schedule: runs tasks on a timer.
  • Projects and workspaces.
  • "White-box" memory: memory the user can read and edit.
  • IM Gateway: connects outside chat apps.

Recent stable releases (GitHub Releases, as of 2026-10-25): v2.2.17 on 2026-09-11, v2.2.18 on 09-20, v2.2.19 on 10-07 and v2.2.20 on 10-18. There are also canary and per-PR nightly builds. v2.2.19 merged 547 PRs from 15 contributors, and v2.2.20 merged 419 PRs from 14.

The default branch is canary, with 1,172 open issues and PRs combined as of 2026-10-25. The code is TypeScript on Next.js and Postgres. GitHub reports the license as NOASSERTION.

LobeHub organizes your agents into 7×24 operation. It hires, schedules, reports on your entire AI team. You stay in charge — without staying online.

LobeHub README (as of 2026-10-25)

Core capabilities

The v2.2.19 and v2.2.20 release notes list these capabilities (vendor claims):

  • Goals: v2.2.19 added a delivery tab, asks for clarification before work starts and raises the main agent's default from 12 to 50 turns. v2.2.20 added batch rollout, widgets and dashboards.
  • Devices: v2.2.19 can expose a device's listening ports and tunnel HTTP and WebSocket traffic. v2.2.20 relays LLM calls through the user's own device and adds remote terminals. The macOS app bundles Computer Use.
  • Outside agents: Codex, Kimi Code, Qoder, Droid and OpenCode can plug in, and so can a DeepSeek Harness adapter.
  • Local-first: topics, projects, agent lists and task boards now read from a local replica.
  • Desktop: the app is now a signed shell around an over-the-air updatable core, so updates no longer need a full reinstall.
  • GitHub loop: a GitHub App wakes the agent that opened a PR when CI fails or a review lands.
  • Recycle bin.
  • Memory and forks: v2.2.20 builds memory from the user's own materials and can fork a chat from any reply.

How it compares

Compared with two similar tools, as of 2026-10-25:

Open WebUI is also a self-hosted web app:

  • As of 2026-10-26, 156,757 GitHub stars; latest GitHub release v0.12.4 (2026-10-25).
  • Its license builds on BSD-3 terms and adds a branding clause. The "Open WebUI" branding may be altered only with 50 or fewer end users in any rolling 30 days, with written permission, or with an enterprise license.
  • Its enterprise docs say internal use with the original branding is free. Rebranding, white-labeling and enterprise-only features need an enterprise license, arranged through sales.
  • The same docs list integration with LDAP, Active Directory and SSO. It is built around giving a group one shared internal chat entry point.

LobeHub's edge is agent orchestration, skills and design. The licenses bite in different places: Open WebUI restricts rebranding, while LobeHub restricts distributing derivatives.

Security advisories. Both have shipped code-execution bugs; LobeHub's GHSA-2g9j was published only in the repo, not in GitHub's global advisory database.

Cherry Studio is a native desktop app:

  • Its README lists Windows, macOS and Linux, plus iOS and Android mobile apps. As of 2026-10-26, the latest GitHub release is v2.1.7 (2026-10-19).
  • The community edition is AGPL-3.0. The README allows commercial use under full AGPL compliance, and sells a separate commercial license for an exemption.
  • Chat history and knowledge bases stay on the device, with local and WebDAV backup options.

It is quick to start with for one person. A team that needs one shared entry point with central accounts is better served by a server-hosted web app such as Open WebUI or LobeHub. LobeHub has a desktop app too, but its core is a server-backed web app with a hosted cloud next to it.

Bottom line: the desktop app is aimed at individual use. For teams that care about identity management, Open WebUI's account integration is more complete. LobeHub's strengths are agents, schedules and a managed cloud, and its license and security record need a separate review.

Who it's for / Not for

Who it's for

  • Individuals who want a polished multi-model AI workspace without running a server. The free cloud tier gives 500,000 credits a month, and users can bring their own API keys.
  • Power users who want to try agent teams, scheduled tasks, skills and MCP plugins, and can live with a product that keeps moving.
  • Students and education institutions, who can get 50% off a cloud plan by applying from a school address with proof of status.
  • Small teams with ops skills that want to self-host for internal use without shipping a modified product.

Not for

  • Companies that plan to modify, rebrand and then distribute or sell it. The LICENSE requires a commercial license first.
  • Organizations whose legal review requires a standard OSI license such as plain Apache 2.0 or MIT.
  • Teams with nobody to track upgrades. The repo's public advisories include remote code execution and auth bypass bugs.
  • Solo users who just want local chat. A desktop app is lighter.

Getting started

Cloud. The cloud edition starts with a sign-up at lobehub.com. On 2026-10-25 the pricing page offered a free trial of MiniMax M3.1 Flash with no credit card.

Self-hosting. The docs recommend Docker Compose. The one-line installer is bash <(curl -fsSL https://lobe.li/setup.sh) -l en. It has local, port and domain modes, serves on port 3210 and runs on Linux or macOS; Windows needs WSL 2. It needs ports 3210, 9000, 9001, 8787 and 8788 free.

Requirements:

  • PostgreSQL 14 or later. The Docker guide uses a paradedb image with pgvector.
  • S3-compatible storage for files and knowledge bases. Redis is optional.
  • At least one model provider key.
  • Secrets such as KEY_VAULTS_SECRET and AUTH_SECRET.

Platforms. The docs warn that Vercel caps functions at 10 seconds and lacks WebSocket support, which is why they recommend a self-managed server; backups and monitoring on that server fall to the operator.

Auth. The options are open access (single user only), Better Auth and a reverse proxy.

Upgrading. The new compose file adds a gateway service. Per the docs, .env has to be updated first, or the container keeps restarting.

Pricing & license

From lobehub.com/pricing, as of 2026-10-25, in US dollars.

LobeHub Cloud

  • Free: $0, 500,000 credits a month, 15 MB of file storage, 100 vector entries.
  • Starter: $12.9 monthly or $118.8 yearly ($9.9 a month), 5M credits, 1 GB of files, 5,000 vector entries.
  • Premium: $24.9 monthly or $238.8 yearly ($19.9 a month), 15M credits, 2 GB, 10,000 vector entries.
  • Ultimate: $49.9 monthly or $478.8 yearly ($39.9 a month), 35M credits, 4 GB, 20,000 vector entries.

How the plans work:

  • Agent Memory starts at Starter; it isn't in the Free feature list.
  • Credits map to tokens, at different rates per model.
  • On Free, the page estimates about 70 messages with GLM-5.3-Flash. GPT-6 Astra, Claude Opus 5.5 and Kimi K3.1 Pro require a subscription.
  • Starter covers roughly 30 Claude Opus 5.5 messages or 10 GPT-6 Astra messages (vendor estimate).
  • Credits reset monthly. Top-up packs are available, and users can add their own API keys.
  • Cancellation goes through Stripe, and the account drops to Free when the plan ends.
  • Students and education institutions can ask hello@lobehub.com for a 50% code. The request must come from a school or institution address with valid proof of status.

Enterprise covers private deployment, a commercial license, branding, user management and private models. It goes through sales.

Cloud vs self-hosting. Self-hosting costs nothing for the software. The server, database, object storage and model APIs are still paid for by the operator, who also handles backups, upgrades and patches. The cloud removes the ops work in exchange for credit billing and data held by the vendor.

License. Going by the LICENSE text itself, it's the LobeHub Community License, copyright LobeHub LLC. It is "based on Apache License 2.0" with two added sets of conditions:

  • Commercial use: commercial use, including as a front-end or back-end service, is allowed as long as the source is unmodified. A commercial license is required to "develop and distribute a derivative work".
  • Contributors: contributors agree that the maintainer may make the license stricter or looser, and that their code may be used commercially, including in the cloud edition.

That makes it a source-available license with commercial strings, not a plain open-source one. GitHub can't match it to a standard license.

What the blog says (vendor's own account). The 1.0 announcement on 2024-06-19 says non-commercial secondary development, including internal corporate use and public services, stays free, and only commercializing a modified version needs a license. That is broader than the LICENSE, but a blog post isn't the license text, and the LICENSE governs where they differ.

License history. The LICENSE commit history shows MIT before 1.0, then Apache 2.0 plus conditions from 2024-06-17, then the rename to LobeHub Community License on 2025-09-21.

Risks & limits

Mixed license messaging. The README's top badge says "apache 2.0", while the License section at the bottom of the same README names the LobeHub Community License. The pricing FAQ says LobeHub "always adheres to the open-source concept". The LICENSE text adds commercial conditions, and it governs. The contributor clause also lets the vendor tighten terms later.

Security record. As of 2026-10-25 every repo advisory was patched before v2.2.20. Affected ranges and fix versions come from the repo advisories; GitHub's global database records some of them differently. Severity ratings for the same bug also differ by source:

  • CVE-2026-23733 (GHSA-4gpc-rhpj-9443):
  • GitHub: critical, 9.6.
  • Advisory: stored XSS in Mermaid diagrams escalates to RCE through the desktop bridge.
  • NVD: shows 6.4 (medium), a score submitted by GitHub as the CNA. The record is deferred.
  • Repo: affects v2.0.0-next.179 and earlier, fixed in v2.0.0-next.180. The global database lists @lobehub/chat 1.143.2 and earlier, with no patched version.
  • CVE-2026-54157 (GHSA-xmwj-c75x-6346):
  • GitHub: medium with no score on the repo page, but critical 9.0 in the global database.
  • Advisory: an unauthenticated /webapi/proxy on app.lobehub.com let anyone proxy requests and plant cookies on lobehub.com.
  • Affects 2.1.56 and earlier, fixed in 2.1.57.
  • CVE-2024-47066 (GHSA-3fc8-2r3f-8wrg):
  • GitHub: critical on the repo page. The global database says medium while still showing a CVSS 3.1 score of 9.0.
  • Advisory: an incomplete SSRF fix that redirects can bypass.
  • Fixed in 1.19.13.
  • CVE-2026-39411 (GHSA-5mwj-v5jw-5c97):
  • GitHub: medium, 5.0.
  • Advisory: "This is an unauthenticated authentication bypass", because the X-lobe-chat-auth request header is only XOR-obfuscated and can be forged.
  • NVD: 7.1 (high).
  • Fixed in 2.1.48.

Other advisories that lead to code execution, all rated medium by GitHub:

  • CVE-2026-42045 (GHSA-xq4x-622m-q8fq): requires the desktop (Electron) app. An attacker gets the model to emit malicious tags for XSS, then calls a command-running interface the desktop main process exposes; the proof of concept uses an attacker-run model gateway. The repo lists 2.1.47 and earlier, fixed in 2.1.48. The global database lists 2.1.26 and earlier, with no patched version.
  • CVE-2025-59417 (GHSA-m79r-r765-5f9j): also desktop-only. XSS in chat messages escalates to code execution on the user's machine. The repo scores it 6.1, fixed in v1.129.4.
  • GHSA-2g9j-v25c-4j97: a path traversal in file names allows arbitrary file writes and code execution. Fixed in 2.1.49, with no CVE.

Fast change. The 2.0 rename changed the product's scope. Old guides and environment variables may break. The V2 upgrade docs cover moving off Clerk or NextAuth.

Setup weight. The server edition needs Postgres, S3 and secrets. On Vercel alone the 10-second function cap applies.

Enterprise. Enterprise plans and commercial licenses both go through sales.

Our verdict

LobeHub wins on polish and speed. The interface is good-looking, agents, skills, memory and schedules keep arriving, and a hosted cloud removes the need to run it.

Three points to weigh first:

  • License: it isn't plain Apache 2.0, so shipping a modified version requires a commercial license. The blog reads more generously, but it isn't the license text.
  • Security: past bugs include RCE and auth bypass, and ratings differ across sources. Self-hosting means someone has to track releases.
  • Direction: it's now an agent platform rather than a chat window, heavy for users who only need a chat box.

Individuals can use the free cloud tier. For teams self-hosting with identity needs, Open WebUI is the comparable option; for solo use, a desktop app is lighter.

Why it stands out

LobeHub, formerly LobeChat, began as an AI chat front end that takes visual design as seriously as model support: a client with a plugin system, multi-provider routing, an agent marketplace and a knowledge base, with every panel looking considered. Since version 2.0 in January 2026 it ships as LobeHub and centres on organizing AI agents into teams that keep working together.

Arvin Xu started the project, and LobeHub LLC maintains it. The engineering is modern Next.js with a real design system underneath and careful attention to state, theming, and motion.

It combines multi-provider routing, plugins, a knowledge base and an agent marketplace, and since 2.0 organises agents into teams.

Sources

Official

  1. GitHub - lobehub/lobehub: 🤯 LobeHub is your Chief Agent Operator, organizing your agents into 7×24 operations by hiring, scheduling, and reporting on your entire AI team.GitHub · Accessed Oct 25, 2026
  2. lobehub/LICENSE at canary · lobehub/lobehubGitHub · Accessed Oct 25, 2026
  3. History for LICENSE - lobehub/lobehubGitHub · Accessed Oct 25, 2026
  4. Releases · lobehub/lobehubGitHub · Accessed Oct 25, 2026
  5. Release 🚀 Release v2.2.20 · lobehub/lobehubGitHub · Accessed Oct 25, 2026
  6. Release 🚀 Release v2.2.19 · lobehub/lobehubGitHub · Accessed Oct 25, 2026
  7. Release LobeHub: Agent teammates that grow with you · lobehub/lobehubGitHub · Accessed Oct 25, 2026
  8. Pricing · LobeHublobehub.com · Accessed Oct 25, 2026

Updates & reading