Automations
Build a workflow Panda runs on a schedule, when something happens, or when you start it.
Feature Updated 18 Nov 2026What an automation is
An automation is a workflow Panda runs for you. That can be a briefing every weekday morning, a summary when a pull request opens, or a note when an email arrives. Use one when the same work comes back on a timetable or in answer to an event, so you stop starting it by hand.
Every automation has a trigger, which decides when it runs, and a flow, the steps it takes each time.
Where to find it
Open Automations in the dock. The list has three tabs: Automations for your own, Webhooks for the addresses other services call, and Templates for ready-made ones. To start a new one, pick one of three cards:
| Card | What it does |
|---|---|
| Describe with AI | You say what you want in plain words and Panda builds the flow. You check it before it runs. |
| Build step by step | You add the trigger and each step yourself in the visual builder. |
| Schedule a prompt | One prompt on a timetable. The builder opens set to every day at 09:00 with one step in place, so you change the time and write the prompt. |
A template asks which project it runs in and what it needs once. Then it opens in the same builder, where any step can be changed.
What it can do
- Run when you choose. On a schedule, when you start it, when something happens in a connected app, or when another service calls its webhook address. One automation can carry up to five triggers, and any of them starts it.
- Take several kinds of step. Ask AI, act in a connected app, branch on a condition, or run a fixed built-in step. An agent step takes a bigger task and works through it on its own.
- Deliver the result. As a notification, a dated file, an update to a file, a new chat in a project, or a note.
- Show every run. The run history lists each run, whether it worked, what each step produced, and an estimate of what it cost.
- Test one part. In the builder, hover a step and choose Run this step or Run up to here. Run in the header runs the whole flow.
Limits
An automation only runs while PandaOS runs on the machine picked under Runs on, in its window or in the background. Unattended runs explains how it keeps going with the window closed, and what it may do without asking you.
An app-event trigger checks the app on a timetable, so it fires at the next check rather than the second the event happens.
Triggers: when automations run
Every automation begins with a trigger that decides when it runs. An automation can carry up to five triggers, and any one of them starts the flow. The same workflow can run on a schedule and also when an email arrives.
| Trigger | When it runs |
|---|---|
| Schedule | On a timetable you pick, such as every hour, every weekday at 9:00, or the first of every month. You set it in a picker, not by writing a cron expression. |
| Manual | Only when you start it. Useful for a client report at the end of a project, or fresh numbers before a meeting. |
| App event | When something happens in a connected app: a new email, a calendar event about to start, a file in a Drive folder, a new pull request. PandaOS checks the app for you and starts the flow when the event matches. |
| Webhook | When another system posts to an address PandaOS makes for this automation. The address contains a secret, stored encrypted on this device. Every step can read what was posted as {{trigger.payload}}, up to 64 KB. |
An app you built yourself can start an automation too. The events it declares appear in the same picker. They fire as soon as the app raises them instead of waiting for the next check.
Apps that can trigger an automation
App-event triggers are available for the apps in this table. Each one has its own events and filters, so a trigger fires only in the cases you care about.
| App | Example events |
|---|---|
| Gmail | New email arrived, email with attachment, email starred, filtered by label, sender or subject |
| Google Calendar | Event starting soon, event ended |
| Google Tasks | Task is due, task added |
| Google Drive | New file added, optionally in one folder or matching a name |
| Google Chat | New message in a space |
| GitHub | New issue opened, new pull request, in a repository you pick |
| Trello | Card added to a list |
| Supabase | SQL condition matched, new advisory finding, new user signed up |
| Database | SQL condition matched on a connected database |
| Vercel | Deployment finished: failed, successful or any |
| HubSpot | New contact, new company, new deal, new ticket, a deal or ticket entering a stage you pick |
You also choose how often PandaOS checks, from every minute up to once a day. Some events have a higher minimum, such as five minutes for a SQL condition. A setting decides what counts as one event:
| Setting | What it does |
|---|---|
| Every match | Fires each time the event matches. |
| Once | Fires one time, then stays off until you reset it. |
| Cooldown | After firing, waits a number of minutes before it can fire again. |
Events that arrive during a cooldown are dropped, not queued. PandaOS remembers which items it has already seen, so restarting the app or waking from sleep never replays a backlog.
The SQL condition triggers for Database and Supabase run your own SQL on a schedule and fire on its result. A trigger fires on any rows returned, a row count greater than a number, or a row count equal to a number. It can also fire on a count crossing a number, once per crossing, or on the first row changing since the last check.
Steps: what each one can do
The flow is a chain of steps, each doing one piece of work. You connect them in order, or branch them with a condition. Each step card shows its kind:
| Step | What it does |
|---|---|
| AI | Reads the output of the step before and writes text, a decision, structured data or a summary. Use it to draft an email, pull key points from a document or sort incoming data. |
| App | Acts in a connected app: sends a Gmail message, creates a Trello card, inserts a Supabase row. It uses the apps you connected in PandaOS. |
| Agent | Runs a task that needs several steps of its own, such as researching a topic and refining a draft. It has its own prompt, model and tools, and does not use one of your agents. |
| Condition | Sends the flow down one path or another based on an earlier result. It branches on a plain expression or asks the model to judge the earlier output, so you need no code. |
| Output | Delivers the result. See the destinations table. |
| Code | A built-in step, listed under Built-in Steps in the step picker. It runs fixed code, so it does exactly the same thing every run. |
| App action | Runs one action of an app you built yourself, with no model involved. Offered only in an automation that belongs to that app. |
An Output step can send the result to:
| Destination | What you get |
|---|---|
| Notification (toast) | A short notification on screen. |
| Date-prefixed file in folder | A new file named YYYY-MM-DD-name.md in a folder you pick. |
| Update a specific file | The run appends to or replaces an existing file. |
| New chat in a project | The run lands as a conversation you can pick up and continue. |
| Note | A markdown note in Notes, bound to the automation's project, another project or none. |
An optional template decides what gets written. Without one, the collected run output is used.
The built-in steps fetch Sentry issues and their latest events, render a triage report, pick fixes for a run and skip known noise. Other built-in steps prepare a worktree per fix, commit the fixes, push them and open pull requests. One more emails the text of an earlier step to your own Gmail address. Each has its own form. A field that needs another step's output lets you pick that step.
In an automation that belongs to an app you built, the step picker shows that app's actions grouped as Does something and Provides. An AI Step and Send the result sit next to them.
Run history and cost
Every automation keeps a run history. Each entry shows when it ran, whether it worked, and the output of every step. When a step failed, you see where and why, including the tool calls it tried and any partial output it produced first.
Each run and each step also shows an estimate of what it cost, because AI steps and app actions use your plan. Watch it on an automation that runs often, and change the expensive step if it costs more than you expected.
Automations use the same skills and connected apps you use in chat. What Panda can do in a chat it can also do on a schedule.