Agent Hub

• Agent Hub is Guild's marketplace and catalog for open-source, community-built AI agents. Users discover agents, install them into a workspace, and run them on Guild's runtime.
• The Hub is designed as an ecosystem: teams do not have to build every agent from scratch when a well-maintained community version already exists.
• Once installed into your workspace, a Hub agent behaves like any other Guild agent — same governance, same sessions, same versioning, same Insights.
• The Hub complements Guild's other agent-creation paths (TypeScript SDK, Goose recipes, Guild Native Markdown). Install from the Hub when a good agent exists; build from scratch when it does not.

Agent Hub is where Guild users go to find agents built by other people rather than build every agent themselves. It is a browsable catalog: agents grouped by use case, searchable by capability, viewable with their descriptions and instructions in the open, and installable into a workspace with a click.

The Hub sits alongside Guild's other agent-creation paths. If you want to write an agent from scratch you can, in TypeScript, in a Goose recipe, or in Guild Native Markdown. If someone has already built the agent you need — a code-review assistant, a release notes writer, a ticket triage bot — the Hub is where you go instead.

The design premise is that the interesting engineering in most agents is not the mechanics. It is the prompt, the tools, the parameters. Once someone has done that work well, the whole ecosystem benefits from installing rather than reinventing.

Browse the catalog

The Hub organizes agents by use case, capability, and popularity. Every listing shows the agent's description, its purpose, and the type it was built with (TypeScript, Goose recipe, or Guild Native). Because agents on the Hub are open, you can see how they are constructed before you install.

Install into a workspace

Installing an agent brings a specific published version of it into a workspace you choose. From that point on, the agent lives inside that workspace and follows the same rules as any other agent there: workspace context applies, workspace credentials scope its tools, and every run generates a session.

Fork to customize

guild agent init --fork owner~agent-name

If an installed agent almost fits your team but not quite, fork it. Forking creates a copy under your account that you can modify, publish, and run as your own version. This is the standard path from "a Hub agent I like" to "a Hub agent I've made my own."

Updates are opt-in

When the author of a Hub agent publishes a new version, installed instances do not auto-update. You choose when to move to a newer version, which keeps you in control of behavior changes even when the source agent evolves.

Not every agent is worth writing yourself

The same agents get built over and over in every organization: a PR reviewer, a standup summarizer, a support-ticket triager. The Hub means teams can start from a working version rather than reinventing the same wheel. That is time back for the agents that actually differentiate your business.

Open agents are easier to trust

A Hub agent's instructions, its tools, its parameters are all visible before you install. You can read the prompt, see which services it will call, and make an informed decision about whether it fits your governance requirements. That transparency is a governance advantage over black-box agents downloaded from anywhere else.

Fork and improve is a real ecosystem

The best Hub agents get forked, improved, and republished. That is how an ecosystem gets better: individual contributions compound into better defaults for everyone. Guild's fork mechanism makes that lifecycle a first-class operation.

Installed agents are still your agents

An installed Hub agent lives in your workspace, uses your credentials, and its sessions belong to you. Nothing about installation gives the original author ongoing access to what the agent does under your roof. Governance and audit trail are entirely on your side.

Read before you install

The Hub's biggest advantage is that agents are open — read the prompt and tool declarations before installing. An agent that looks useful in the description but does surprising things in its instructions is safer to catch at browse time than at run time.

Credentials are still on you

An installed agent that calls GitHub still needs GitHub credentials configured in your workspace. Nothing about installing an agent connects services for you. Check the agent's required tools against the credentials your workspace has connected.

Forking is often the right first move

Once you install a Hub agent, forking gives you a version under your account that you control end to end. If you plan to modify the agent at all, or if you want to pin behavior for compliance reasons, fork it rather than run someone else's published version indefinitely.

Contribute back if you improve it

A better version of an existing Hub agent helps everyone. If your team fixes a bug or improves a prompt in a forked Hub agent, consider publishing it publicly on the Hub so others benefit. The ecosystem only compounds if people share the improvements.

Guild is a control plane for AI agents, a place to build, deploy, and govern the agents your teams run in production. Agent Hub is where that ecosystem starts — a catalog of agents built by others, installable into your workspace with the same governance every other Guild agent gets.

• Explore the Agent Hub landing page for the full pitch.
• Once installed, agents live inside a workspace and generate sessions like any other Guild agent.
• Sign in at app.guild.ai to browse the Hub and install your first agent.