Perplexity has turned its Computer agent into something closer to an always-on co-worker. The company's new Automations feature lets Perplexity Computer run ongoing jobs on a schedule or in response to events in connected apps such as Slack, Gmail, Outlook, Linear and GitHub, remembering what it did on earlier runs.
Launched this week, Automations replace Perplexity's older Scheduled Tasks and are available to all Computer users. This guide covers how the feature works, how to set up your first automation, and how to stop an event-triggered agent from causing problems.
How Automations Work
According to Perplexity, each automation combines four parts:
- Instructions – what you want done, written in plain language
- A trigger – a schedule, an event from a connected app, or a check for missed updates
- Connected resources – your apps, files, Projects, memory and skills
- Run history – every earlier run, which the agent can consult before acting
That last part is what makes Automations different from a basic cron job. Each run carries memory of previous runs, so the agent builds on its past work instead of starting from scratch each time.
There are three trigger types:
- Scheduled – fixed times, such as "every Monday at 8 a.m."
- Event-triggered – activity in a connected app, narrowed with conditional filters
- Missed-update checks – watching for overdue replies, stalled tickets or missed deadlines
Pricing is simple: credits are spent only when an automation actually runs, not while it is waiting for a trigger.
Setting Up Your First Automation
1. Start from a clear, repeatable job
Pick work you already do on a schedule or in response to a signal, such as a weekly status summary or replies to a certain type of email. Repeatable, well-bounded tasks are the best fit.
2. Describe it in plain language
Open the Automations dashboard in Perplexity Computer and describe the job. For schedule-based work, Computer suggests a schedule for you to confirm. Be specific about the output you want: a draft, a summary, a pull request or a Slack message.
3. Choose and narrow the trigger
For event triggers, pick the connector and add conditions so only relevant events start a run. For example, Gmail messages from a specific customer domain, or Linear issues with a particular label. Tight filters save credits and reduce risk.
4. Connect only the resources it needs
Give the automation the minimum set of apps, files and Projects needed for the job. An automation that drafts customer replies does not need write access to your code repository.
5. Test, review and iterate
Use Run now to test before relying on the trigger. Open the automation to check its instructions, status, run history and outputs, then edit or pause it as needed. If you used Scheduled Tasks, your existing jobs appear in the new panel and migrate when you open them.
Three Practical Examples
Perplexity's launch materials point to patterns that suit most teams:
- Customer replies: watch Gmail for messages from key accounts, check earlier exchanges, and draft responses for your approval
- Engineering changes: when a Linear issue gets a specific label, implement the change and open a pull request for review
- Decision tracking: update a decision log from Slack discussions, mark superseded decisions, and flag tasks they affect
In each case the agent prepares work and a human approves it. That is the right default for any new automation.
Keeping Event-Triggered Agents Safe
Event triggers carry a specific risk. Before launch, RuntimeWire pointed out that letting an incoming email start an agent workflow means untrusted content can start your agent. A crafted message could try to prompt-inject instructions into the run. To protect yourself:
- Prefer drafts over sends. Have the automation produce drafts, comments or pull requests rather than acting directly.
- Filter senders and sources. Use conditions to limit triggers to known domains, channels or labels.
- Separate read and write access. Give monitoring automations read-only access wherever possible.
- Review run history weekly. Look for odd actions, unexpected tool use or outputs that drift from the instructions.
- Pause before you leave. If you will be unavailable to review outputs, pause sensitive automations.
Why It Matters
Automations put Perplexity in direct competition with this week's other always-on agent launches, including OpenAI's dots and Microsoft's Copilot Autopilot. Paying only per run, and the clear structure of instructions, trigger and history, make it easy to experiment cheaply. The memory between runs is the real advance: an agent that remembers last week's triage can improve over time instead of making the same mistakes.
Start small. Automate one low-risk report, keep approvals on, and expand only after the run history shows the agent consistently doing what you intended.
