Code
TypeScript that runs in an isolated VM. Fast, low cost, and the same result each time. Use it for fixed rules, filters, and data changes.
Managed Agent Automations
A managed agent automation is a workflow that Type runs for a Space. A trigger starts each run. You do not keep a server, a queue, or a scheduler. Type runs the work, keeps the history, and repairs or pauses the automation when a run fails.
Each automation comes with a managed agent: Claude or Codex in the Space, with the Space’s connections, skills, and memory. Each run can use four things, alone or together.
Skills
Reusable instructions and scripts that your team saves in the Space. One skill can serve many automations.
A managed agent
Claude or Codex, with the Space’s connections, skills, and memory. Use it for research, writing, and decisions that need judgment.
A sandbox
An isolated Linux machine. Run shell commands, Git, and the GitHub CLI. Files can stay between runs.
All together
The strongest automations use all four. Code filters and acts, the sandbox runs commands, and the managed agent follows a skill for the steps that need judgment. Combine them all.
What starts a run
Section titled “What starts a run”| Trigger | Starts a run when | Instructions | Code |
|---|---|---|---|
| Schedule | A time comes: every few minutes, daily, monthly, or a cron expression | Yes | Yes |
| Slack | Someone posts, mentions Type, or adds a reaction | Yes | Yes |
| GitHub | A pull request, issue, push, review, release, or CI run changes | Yes | Yes |
| Linear | An issue, comment, project, cycle, label, or reaction changes | Yes | No |
| Webhook | Another service sends an HTTP event | Yes | Yes, with an adapter |
| Email and RSS | Someone sends an email to the Space, or a feed has a new item | Yes | No |
Two ways to build
Section titled “Two ways to build”Instruction automations. Write what you want in plain words. On each run, the managed agent reads your instructions and uses the Space’s skills, connections, and sandbox to do the work. Make one in the Space, or ask Type to make it for you.
Code automations. Write TypeScript. Your code reads events, filters them, calls connected apps, and runs commands in the sandbox. When a step needs judgment, the code gives that step to the managed agent. Deploy with Type CLI, or ask Type to write and deploy it.
| Instruction automation | Code automation | |
|---|---|---|
| You write | Instructions in plain words | TypeScript |
| Each run | The managed agent follows the instructions. | Your code runs. It can call the sandbox and give steps to the managed agent. |
| Uses skills | The managed agent uses the skills that you name, and others that fit. | Through a handoff to the managed agent |
| Uses the sandbox | The managed agent uses its sandbox. | ctx.tools runs commands in the automation’s sandbox. |
| Cost | AI usage on each run | Low. AI usage only for handoffs and AI filters. |
| Result | Can change from run to run | The same each time |
| Status | Available | Early access |
Examples
Section titled “Examples”| Automation | Trigger | What runs |
|---|---|---|
| Morning support digest | Schedule, 8:00 each weekday | A managed agent reads new tickets, uses the team’s triage skill, and posts a summary. |
| Deploy failure triage | Slack message that contains “deploy failed” | Code filters the message and replies with the runbook. The managed agent finds the cause and posts it in the thread. |
| Merged pull request notes | GitHub pull_request.closed, merged into main | Code reads the pull request with the GitHub CLI in the sandbox and posts release notes. |
| CI failure review | GitHub workflow_run.completed | An AI question in the matcher skips known flaky jobs. The managed agent reviews the rest. |
| Meeting log | Webhook from a meeting service | A webhook adapter verifies the request. Code saves the meeting. The managed agent writes action items. |
What Type does for you
Section titled “What Type does for you”- Runs the triggers. Type keeps schedules, receives events, and filters them before your work starts.
- Gives access safely. Each run uses the Space’s connections. Code never sees a credential.
- Keeps history. Each run records its output, its steps, and its errors.
- Handles failures. A failed run pauses the automation. For code automations, Type starts a repair agent that fixes the code and tests it.
Reports
Section titled “Reports”When automation reports are on for your workspace, Type can save a chart, a table, or a summary with an automation. Ask for a report when it helps you understand the results. An automation that does not need a report shows its activity, with no empty report.
- Explore a report to filter, sort, or expand its data.
- To change a report, select a part of it and describe the change in the chat next to it. A change to the presentation does not run the automation.
- To hide a report at once, ask Type to remove it.
- Type sends a report to another place only when you ask. Slack gets a version made for Slack messages, with images when necessary. Email gets a static report. You can explore a report only in Type.
In a code automation, save a report with ctx.setVisualization(), and a one-sentence result with ctx.setSummary().
Next steps
Section titled “Next steps” Quickstart Make an instruction automation, then a code automation.
How automations work Triggers, runs, access, state, and guarantees.
Run a managed agent Instructions, models, skills, and connected accounts.
Run code Write a code automation with the Type SDK.