lobe-chat-agents: The Community Agent Index Behind LobeChat's Marketplace

What lobe-chat-agents Is and How LobeChat Uses It

LobeChat is an open-source chat application that supports loading custom AI agents from a marketplace. The lobe-chat-agents repository is the source of that marketplace data. LobeChat accesses the built index.json file from this repository to display the list of available agents to users. The index is not a static file checked in once; it is regenerated by the build scripts whenever a pull request merges a new agent, ensuring the marketplace reflects current submissions.

The repository serves two separate audiences. Developers who build and publish agents interact with the src/ directory, the agent template files, and the pull request process. LobeChat's application code fetches the published index at runtime. End users of LobeChat never interact with this repository directly; they see the agents as a browsable list inside the chat application, accessible at lobechat.com/discover/assistants.

Repository Layout: src/, Templates, and the Build Output

The top-level repository structure is organized around contribution and build:

- src/ holds one JSON file per submitted agent. Each file defines the agent's system prompt, display name, description, tags, and configuration. - agent-template.json is the minimal template for a new submission, with only the required fields. - agent-template-full.json is a comprehensive template listing every supported parameter. The README notes that not all agents will be accepted, and recommends choosing parameters from the full template based on specific needs. - schema/ holds the JSON schemas that validate submitted agent files. - scripts/ holds the TypeScript build and validation scripts run in CI (build.ts, format.ts, validate-language.ts, test.ts, and auto-submit.ts). - locales/ holds the i18n translations. - public/ holds the built output including the index.json that LobeChat fetches.

The package.json names the package @lobehub/agents-index at version 1.42.0. The build system uses Bun (bun scripts/commands/build.ts).

How to Submit an Agent to the Index

The README documents a five-step pull request process:

1. Fork the repository. 2. Copy agent-template.json or agent-template-full.json. 3. Fill in the copy and rename it to match the agent. 4. Move the file into the src/ directory. 5. Open a pull request and wait for review.

The README includes two notes. First, the createAt date field is populated automatically after merge and should not be set manually. Second, not every submitted agent is accepted; the maintainers review each submission. Agents can be submitted by someone other than the author, but the README states that author submissions are preferred. If an existing agent needs its description corrected or should be removed, the process is an issue or pull request.

The README also links to a direct submission interface at the project website for users who prefer not to work with the repository directly.

What Agents the Index Currently Covers

The README lists a sample of the agents in the index at the time of writing. The examples span a wide range: a lateral thinking puzzle host, a food reviewer, an academic writing assistant, a Minecraft senior developer, an open source license analyst, a Python and VS Code expert, an academic paper reading mentor, a nutritional advisor, a recipe assistant, a YouTube summarizer, and a green plant care expert, among others. The range of categories represented in the sample alone (games, food, academic research, software development, health, and content analysis) illustrates that the index is not restricted to a single professional domain.

Each entry in the index includes the agent name, the author's GitHub handle, the submission date, a short description of the agent's specialty, and a set of tags. For the Python VS Code expert, those tags are: python, vs-code, programming, ai-assistant, development. The tags serve as the primary axis for filtering and discovery inside the LobeChat marketplace. The README's agent listing is a sampled excerpt from the full index; the complete browsable list is at lobechat.com/discover/assistants. The package.json version (1.42.0) reflects how many versioned releases the index has had since the repository was created.

Self-Hosting and Local Development

The README documents two options for running the index outside of the main lobechat.com deployment. The Deploy to Vercel option links to one-click Vercel deployment for teams that want to host their own agent marketplace separate from the public one, for example a corporate deployment that restricts which agents employees can access. The local development section covers running the index locally.

For local development, the build script uses Bun. The package.json scripts include build (runs build.ts), test (runs test.ts), i18n:validate (validates language files), and format (formats agent files). The release script uses semantic-release, and changelog generation is included in the repository root as CHANGELOG.md.

What the Index Does Not Include

lobe-chat-agents is a marketplace index, not a development framework. It does not contain the LobeChat application code, the LLM API integration, or the conversation management logic. Those live in the main lobehub/lobe-chat repository. Agents in this index are configured AI assistants with specific system prompts and settings; they are not plugins or tool-use integrations in the sense of running external code. A developer who wants to publish a LobeChat plugin with custom API calls needs the plugin development documentation in the main repository, not this index.

The repository also does not handle agent versioning within the index. Once an agent is submitted and merged, its entry in src/ is the current version. There is no mechanism documented in the README for maintaining multiple versions of the same agent. Updating an existing agent requires a new pull request modifying the agent's file in src/. The release history visible in CHANGELOG.md tracks index releases (v1.42.0 in package.json) rather than individual agent versions.

Compared to OpenAI's GPT Store

OpenAI's GPT Store is a closed marketplace where users publish and discover custom GPT configurations. Submission goes through OpenAI's hosted platform, and published GPTs are accessible only within ChatGPT's interface. Configuration is stored on OpenAI's servers.

lobe-chat-agents takes the opposite approach: the entire index is an open, version-controlled repository on GitHub. Every submitted agent is a readable JSON file in src/. Any team that wants to fork the index and run their own private marketplace can deploy the built index.json to their own server (or Vercel) and configure a self-hosted LobeChat instance to read from it instead of the public URL. That openness also means the review process is public and the full submission history is visible in the repository's pull request log.

Editorial conclusion

Developers who have built a custom LobeChat agent and want to share it with the broader LobeChat user base should find the pull-request submission process in this repository the right path. Individual users who want to use LobeChat agents do not interact with this repository directly: they browse and install agents through the LobeChat interface. Before submitting, read agent-template-full.json carefully for the full set of configurable parameters, confirm your agent tags match the existing taxonomy in the src/ directory to help the review team, and note that not every submission is accepted.

Frequently asked questions

Official sources

1. License: MIT
2. lobehub/lobe-chat-agents on GitHub
3. Project website
4. README
5. Releases