Create an automation

Automations are saved agent workflows that run on a schedule or on demand. Manage them from the Automations workspace in the left sidebar.

note

Schedules fire only while Glean Tau is running. There's no background or cloud execution, so keep the app open for scheduled runs.

Each automation is a reusable task definition plus its run history. When Glean Tau runs an automation, it starts a session in a selected project folder, follows the saved instructions, and records the result as an unread run entry with a status and summary. You can open any completed run like a normal chat to see the transcript, tool calls, and evidence.

Each automation definition has the following parts:

Setting	What it controls
Name	Title displayed in Automations workspace
Instructions	The prompt the agent runs
Project and folders	The local folders each run inspects
Access mode	Read-only or Full access (can write and run commands). This bounds what a run can do, alongside any extension-level limits such as Agent browser controls.
Reuse chat	Whether each run continues the previous automation chat or starts a fresh one. On by default, so runs build on the same chat.
Model and reasoning	Optional overrides per automation
Schedule and timezone	Optional cron expression with timezone. Schedules are the only automatic trigger. There are no event, file-watch, or webhook triggers.
Run if missed	Whether to queue a run after missing a scheduled run. A stale scheduled time is skipped by default.
Memory	Whether to inject summaries from recent runs

Decide when to use automations

Use automation when the work is recurring and bounded, and when a report is a useful result. Good fits include:

• Standup summaries and daily planning
• Weekly release notes from merged pull requests
• Pull request triage and CI failure monitoring
• Dependency sweeps and repository health checks
• Test-gap scans on recent changes

Avoid automations for ambiguous product decisions, destructive actions, broad unsupervised rewrites, or anything that would expose secrets to a recurring run.

An automation can't stop to ask a clarifying question, so its instructions must be self-contained. When something is unclear, tell it to state its assumptions or report what's missing.

Create an automation

1

Open the Automations workspace

Open Automations from the left sidebar and select New automation. Choose Create with Glean to describe the automation in a chat, or Set up manually to fill in the form yourself.

2

Add instructions and folders

Enter the instructions and select the project folder or folders.

3

Set the access mode

Under Advanced options, set the access mode. Full access is on by default when your organization allows it, so turn it off for read-only runs.

4

Set the run options and overrides

Also under Advanced options:

• Run memory, on by default, includes summaries of previous runs as context.
• Reuse chat, on by default, keeps every run in one run chat. Turn it off to start a fresh run chat each time. This setting controls how runs relate to each other, not whether a run reuses your build chat.
• Catch up runs, off by default, runs once to catch up on a missed schedule instead of skipping it.

Set the model and reasoning if you want to override the defaults.

5

Add a schedule

Add a schedule and timezone if the automation should run automatically.

You can also manage automations from any chat in natural language. The automation tool supports create, update, list, get, run now, and delete. New automations created this way are active right away, and they use Full access unless you specify Read-only.

Understand the chats an automation uses

An automation uses separate chats for building and for running. Selecting Run now does not continue the chat you built the automation in. It starts a run in a separate chat.

• The build chat is where you describe the automation after you select Create with Glean. It ends when you save the automation.
• A run chat is where the automation does its work. Every run reports in a run chat, never in the build chat.

Both kinds of chat stay in your chat list. Run chats carry an Automation badge, so you can tell them apart. The build chat from Create with Glean isn't attached to a project, so look for it in your full chat list rather than under the project you were working in.

Because the build chat stops at the point of saving, don't rely on it for run history or screenshots. Use the automation's run history instead.

Run and test an automation

Open Automations, select the automation, then select Run now.

• New automations are active as soon as you create them, so a schedule starts right away. To test an automation first, create it without a schedule or pause it, select Run now, and review the output. Then add the schedule.
• Scheduled runs use a five-part cron expression and the selected timezone.

An automation with several folders produces one run per folder.

Change an existing automation

Select the automation, then select Edit. Glean Tau opens a new chat that already knows which automation you're changing, so you can describe what to adjust.

You can also ask in any chat to update an automation by name.

Write a schedule

You don't have to write a cron expression yourself. Ask in a chat for the timing you want, such as "run this every weekday at 9 AM," and Glean Tau sets the schedule for you.

If you enter an expression by hand, Glean Tau describes it in plain language as you type, such as "Weekdays at 9 AM." Check that line before you save. If it doesn't appear, the expression is either invalid or too complex to summarize.

Schedules follow the timezone you select, not the timezone of your computer. A run fires only while Glean Tau is open.

A schedule is a cron expression with five fields, separated by spaces. Each field sets one unit of time:

*  *  *  *  *
│  │  │  │  └─ day of week (0-7, where 0 and 7 are both Sunday)
│  │  │  └──── month (1-12)
│  │  └─────── day of month (1-31)
│  └────────── hour (0-23)
└───────────── minute (0-59)

Each field takes a number or one of these:

Syntax	Meaning	Example
*	Every value	* * * * * runs every minute
a-b	A range	0 9-17 * * * runs hourly from 9 AM to 5 PM
a,b	A list	0 9 * * 1,4 runs Monday and Thursday at 9 AM
*/n	Every n units	*/15 * * * * runs every 15 minutes

Common schedules:

Expression	When it runs
0 9 * * *	Every day at 9 AM
0 9 * * 1-5	Weekdays at 9 AM
30 17 * * 5	Fridays at 5:30 PM
0 */4 * * *	Every 4 hours
0 9 1 * *	The first day of each month at 9 AM

Use numbers, not names. Glean Tau doesn't accept MON or JAN, and it doesn't support shortcuts such as @daily or the L, W, #, and ? characters you may see in other cron tools.

Manage automations

• Use Activate and Pause to control whether an automation is eligible for scheduled runs. An automation missing required setup shows Needs setup.
• Search the list by name, instructions, folder path, or status, or filter it by All, Active, or Paused. Open run history to review past results.
• Use Delete run to remove a finished run you no longer need, and use Delete to remove an automation. You can't delete an automation while it has queued or running runs.

warning

In Full access mode, writes and commands aren't bounded to the selected folder, so treat the automation as trusted code. Review any output before you commit, push, or share it.

Limitations

• Your computer must be awake. Automations run only while your computer is awake and online. Turn on Keep awake in the Automations workspace to keep them on schedule. By default, a schedule missed while the computer was asleep is skipped unless Catch up runs is on.
• Interrupted runs fail. If Glean Tau quits during a queued or running automation, that run is marked failed when the app restarts.
• Concurrency is capped. Up to 10 runs execute at once across all automations. Runs against the same folder are serialized, and Reuse chat serializes all of an automation's runs because they share one session.
• No interactive prompts. A run can't ask clarifying questions, so instructions must be self-contained.
• No self-creation. An automation run can't create other automations.

See also

• Agent approval modes — how access modes gate filesystem, process, and network actions
• Skills — reusable instruction bundles for repeated tasks