Skip to main content
Guild is a control plane for AI agents. Teams use it to build and run agents, and to govern what each one can access and spend: which services and models it may use, and what it may do with them. Agents live in workspaces your team shares. Install a ready-made agent from the Agent Hub, or build your own with the Guild CLI and Agent SDK.

​
What Guild does

  • Runs agents. Agents run in Guild’s runtime, separate from the control plane. Start one from chat, or let a trigger start it from a webhook event, a schedule, or an API call.
  • Keeps credentials out of agent code. Guild stores the credentials for services like GitHub, Slack, and Jira and adds them to each request server-side. Credential policies limit what an agent can do with them, such as which repositories or channels it can touch.
  • Controls models and spend. Agents use Guild-managed model access by default, or your own keys for Anthropic, OpenAI, Google, and others. With your own keys, model policies set which models the account, a workspace, or a single agent may call. Spend budgets cap monthly spend per workspace or agent.
  • Records what agents did. Every session keeps an event log of LLM calls, tool calls, and errors. Organization admins get an audit log, and the usage dashboard tracks spend by workspace, agent, user, and model.
  • Connects to your tools. Built-in integrations cover GitHub, Slack, Jira, Gmail, Google Drive, and more. Add any other HTTP API as a custom integration.
  • Includes a built-in agent. The Smith, Guild’s built-in agent, answers questions, helps build and debug agents, and can commission a Software Factory that turns a labeled issue into a reviewed pull request.
See Security architecture for how the control plane, runtime, and credential proxy fit together.

​
Developer resources

Agent SDK

Build Native, TypeScript, Goose, OpenClaw, or LangGraph agents with tools, typed inputs, and platform services.

Guild CLI

Install the CLI, scaffold an agent, and run it locally before publishing.

MCP server

Call Guild operations as MCP tools from Claude Code, Cursor, or any MCP client.

Triggers & webhooks

Run agents automatically from webhook events or recurring schedules.

Trigger API keys

Authenticate programmatic requests to Guild triggers.

Integrations

Connect agents to GitHub, Slack, Jira, Google Workspace, and more.

​
Core concepts

Workspaces

A workspace is a container for agents, triggers, context, and credential policies.

Agents

Agents are programs that accept input, use tools and LLMs, and produce output. They range from simple scripted programs to autonomous LLM-driven workflows.

Sessions

A session is a log of an agent’s run. Input or a trigger is provided, the agent runs, asks questions if needed, and returns output or achieves an outcome.

​
How agents work

Requests flow from the browser or CLI to Guild’s API, which queries the database and forwards the request to the runtime, which executes the agent.
  1. You start a session and send a message to an agent.
  2. The Guild API receives the request, authenticates it, and forwards it to the runtime.
  3. The runtime loads the agent code, executes it, and manages the agent lifecycle.
  4. The agent can call tools, invoke sub-agents, make LLM calls, and ask you questions.
  5. When the agent finishes, its output is returned to your session.

​
Next steps

Quickstart

Build and publish your first agent in five minutes.

CLI reference

Install the CLI and learn the development workflow.

Agent SDK

Build advanced agents with typed inputs, tools, and platform services.
Last modified on October 15, 2026