Stevin Manual/Features/Automations Draft preview Updated November 19, 2026

Automations

Have Stevin run an instruction on a schedule or when something happens: triggers, what one run does, which tools it may use, limits, and run history.

What an automation is#

An automation is a trigger plus one instruction. At the set time, or when the event happens, Stevin follows the instruction you wrote and delivers the result.

It suits work that repeats, such as a morning briefing, a weekly report, assigning new leads, or checks after a release.

Create an automation#

There are two ways:

  • Say it in chat: "From now on, at 9:00 every weekday morning, summarize the pull requests merged yesterday and post them in this channel." Stevin sets it up for you.
  • Build it in the console: open Automations, create a new one, and fill it in as described below.

Triggers#

An automation can have several triggers. It runs when any one of them is met.

Trigger When it fires Setup
Schedule Every day, every weekday, chosen days of the week, or monthly, at the time you set Pick it in the visual editor, or write a recurrence rule (RRULE) for anything more complex. Times are in your browser's time zone
Feishu or Lark Base record A record is added to the chosen Base table, or changed Needs the Base automation capability on the Feishu or Lark integration. The Base-side workflow is created and removed for you
GitHub event A selected event happens in the chosen repository, such as a new pull request or issue Uses your connected GitHub integration. Nothing to paste on GitHub's side
Webhook Another system calls the automation's own address See the next section

Webhook triggers#

When you save an automation with a webhook trigger, it gets its own address and key. Both appear on the automation's page.

  • Send the key in an Authorization: Bearer header or an X-Webhook-Secret header. A header is the safer choice because it stays out of logs and browser history.
  • For systems whose whole configuration is one URL box, the page also gives an address with the key already in it.
  • Filter: optionally, a regular expression matched against the JSON body of each delivery. Only deliveries that match start a run, so a busy source can be narrowed to the events you want. Leave it empty to run on every delivery.
  • Rotate the key at any time. The address stays the same; give the new key to every sender.

Treat the key as a password: anyone who has it can make the automation run.

The instruction#

Write what one run should do, in plain language, as you would brief a colleague. How often it runs comes from the triggers. Make clear:

  • What to read.
  • How to process it and what to compare or judge by.
  • Where to send the result and in what format.

Run settings#

Setting What it does
MCP servers Which installed MCP servers this automation may use. Select only what it needs. Workspace permissions still apply
Model The model for future runs. By default the workspace's automation model; see Default models
Run limit Caps how many times it may run in a period. A schedule can run uncapped; event-driven automations get a default ceiling so an event storm cannot run away

How runs behave#

  • Each run is a task and appears in Activity under the Automations filter.
  • Actions that need approval still stop and wait for someone to confirm. For an automation to run unattended, the write tools it uses must be set to run directly. See Approvals.
  • Runs over the limit are skipped.

Test runs and history#

  • Run now starts a run immediately with the saved instruction. For an event-driven automation you can add context that the run should take into account.
  • The automation's page shows its recent runs with their outcome, and a chart of the last days.

Pause, change or cancel#

  • You can edit the instruction, triggers and settings at any time; future runs use the new version.
  • Cancelling an automation stops it and takes down any webhook it created.
  • When an integration it depends on is disconnected, its runs fail. They recover once you reconnect.

Advice#

  1. Do the job once by hand in chat first, and turn it into an automation once you are happy with the result.
  2. Write "do not send a message when there is nothing new" in the instruction, to avoid empty messages.
  3. Set a run limit for frequent events.
Last updated November 19, 2026 Suggest an edit