Functions (Beta)

Beta

The authoring surface is defineJob inside Harbor runtime files. Do not add legacy Orbit package imports to new Function docs.

Shape

summarize-message.job.ts

defineJob({
  name: "summarizeMessage",
  description: "Summarize a message for the current workspace.",
  input: {
    message: "string",
    tone: "string?",
  },
  output: {
    summary: "string",
  },
  async run(input) {
    const summary = await hrbr.ai.text({
      model: "openai/gpt-4.1-mini",
      prompt: "Summarize this message in one paragraph: " + input.message,
    })
    return { summary }
  },
})

Required fields:

Field	Purpose
`name`	CamelCase or kebab-case name. Harbor canonicalizes it for published lookup.
`description`	One sentence shown in dashboard and inspection output.
`input`	Static input schema. Shortcuts include `string`, `number`, `boolean`, `unknown`, and optional `?`.
`run`	Inline async function executed by Harbor.

Publish and inspect

Use inspect when you are authoring or checking metadata:

hrbr inspect -f ./summarize-message.job.ts
hrbr inspect 'return await hrbr.jobs.inspect({ name: "summarize-message" })'

Use exec when you want a traced execution run around the publish or the code path:

hrbr exec -f ./summarize-message.job.ts

Run

Run published Functions from the dashboard or from SDK clients. The local CLI is currently the authoring and inspection entrypoint, not the invocation history surface.

Runtime globals

Function code can use Harbor runtime globals such as plugin namespaces and hrbr runtime primitives. Discover exact source/tool names first:

hrbr inspect 'return await hrbr.tools.search({ query: "send slack message", limit: 3 })'

Then use the discovered namespace in the Function body.

@hrbr/client and Python harbor-sdk call the Harbor API. @hrbr/sdk carries system contracts and generated protocol pieces. defineJob is the runtime authoring global used by files submitted through hrbr inspect or hrbr exec.

Your first Function

Walk through publish and inspect.

Runtime execute

Call execution from TypeScript or Python clients.