Skip to content
LogoLogo

Bento

Bento runs coding agents from a definition you keep under version control. The pattern is harness-as-code: the trigger, runtime, skill, sandbox, and guardrails around an agent live in your repository, not in a console. You write that definition as a pipeline in YAML. The daemon watches for a matching event, provisions an isolated workspace, and runs the agent in it.

To install the daemon and write your first pipeline, read Installation.

Triggers and outputs

Bento receives work from the tools around your repository and returns the result to a declared destination. Each card below is a trigger that starts a run, a control that fires one on demand, or an output that receives the result. Open a card to configure it.

TriggerGitHubPull requests, issues, reviews, and repository eventsTriggerLinearIssue events delivered through workspace webhooksTriggerWebhooksNamed HTTP endpoints for events from other systemsTriggerSchedulesRecurring runs declared with UTC cron expressionsControlCLI & MCPFire pipelines on demand or expose them to agentsOutputSlackPost completed pipeline output through a named bot

A concrete example

A pull request opens on GitHub

GitHub sends a webhook to the daemon.

The daemon matches the event to a pipeline

A pr-review pipeline fires. The daemon retrieves the diff and injects relevant context from the knowledge base.

The daemon dispatches an agent to a sandbox

The pr-review agent executes in an ephemeral sandbox with the repository checked out.

The agent posts its findings back

The agent's output lands as a pull request comment.

Next steps

  • Installation — install the daemon and configure your first pipeline
  • Concepts — pipelines, agents, skills, workloads, and sandboxes explained