Automation

Automation is for repeatable work whose inputs, output, and permission boundary can be written down in advance: daily digests, periodic checks, or a one-time reminder. Every trigger creates a fresh General Work session; it does not reuse the temporary context of an older conversation. For one-off to-dos, work that should wait for a person, or continuing an existing thread, use the task board.

Before you begin

• At least one enabled, configured model; the model, workspace, Skill, and MCP must still be available when the run starts
• A prompt that names the inputs, output location, failure behavior, and acceptance checks
• A task that does not depend on a person completing login, CAPTCHA, clarification, or a high-risk approval
• An understanding that Automation is a local desktop scheduler, not a cloud cron service

Open the automation list

Open Automations in the left rail. Scheduled tasks lists each automation’s name, schedule, and next run. The toggle keeps the definition without scheduling. The play control runs it immediately, without waiting for the clock.

Scheduled tasks lists name, schedule, and next run. The toggle controls scheduling; the play control runs once now.

Start from a template

Use Add from template when you do not want to write the prompt from scratch. Templates cover digests, reminders, and routine checklists. Review the prompt, model, and permissions before saving.

A template is a starting point. Review the prompt, model, and permissions before you save.

Fill in an automation

Automations currently use General Work, so there is no Agent selector. Choose a model, then configure the workspace, Skills, MCP, tool approval, and access scope.

1. Enter a name and prompt. State the output format, input scope, output path, prohibited actions, and acceptance checks.
2. Choose a model and thinking depth. The model must remain enabled and configured when the scheduled run starts.
3. Select a workspace for local material; add a Skill or connected MCP only when the task needs it.
4. Choose a recurring schedule, interval, or one-time run. Optionally set an active date range; turn off Enable after saving to save a draft.
5. Save. Missing name, prompt, or model, an invalid schedule, and an invalid date range are shown next to the relevant field before a request is sent.

Enter the name, prompt, model, and workspace, then choose recurring, interval, or once. Full access and bypass approval stay visible.

◇

Approval checkpointFull Access and Bypass tool approval are highlighted because they let background work continue while you are away. They do not bypass system elevation, explicit security blocks, or unavailable resources.

New automations default to Full Access + Bypass tool approval and require confirmation before saving. For a task that only reads and organizes workspace material, switch to Default Access + Auto-approve normal actions, then validate once with Run now.

›_Daily digest example

Every weekday at 09:15, read files updated in the last 24 hours under inbox/ and notes/ in the current workspace. Write a Markdown brief to reports/daily-brief.md with items, source paths, suggested priority, and questions requiring confirmation. Do not delete, overwrite, or move existing files. If there is no new material, write “No new material today” and state the checked scope. Before finishing, verify that every conclusion has a source path.

Schedule, active dates, and recovery

Setting	Use it for	Keep in mind
Recurring	Daily, weekdays, weekly, or monthly at a local clock time	A monthly day that does not exist in a month is skipped
Interval	Every N minutes, hours, or days	Avoid intervals shorter than the task can complete
Once	A single date and time	The task is no longer scheduled after the run is collected
Active dates	Limit when the schedule is eligible	The end date cannot precede the start date

The scheduler is active while the Wordless desktop process is running. When the app starts or the device resumes, schedules are restored and the latest eligible missed run may be recovered; every missed run is not replayed. This is not a cloud SLA or a second-precise job system. Use a server-side scheduler for work that must run while the computer is off.

Tool approval and access

Setting	Runtime behavior	Use it when
Ask every time	Each tool operation waits for your decision	You are still testing or writing important material
Auto-approve normal actions	Ordinary actions continue; high-risk actions pause	A verified low-risk read and organize workflow
Bypass tool approval	Ordinary and high-risk tool operations are approved automatically	The scope is narrow, reversible, and thoroughly tested
Default Access	File tools stay inside the selected workspace	Most local automations
Full Access	A wider file and tool boundary is available	The target paths and impact are understood

With Ask every time, or when Auto-approve reaches a high-risk action, the run becomes waiting until you resolve an approval or user question.

Inspect every run

Open Run history, filter by status, and expand a row. Do not treat completed as proof of a correct result; open the linked session and check the artifact, sources, tool parameters, and errors.

Run history filters by queued, running, waiting, completed, failed, and related states. Expand a row for the completion time and linked session.

Status	Meaning	Next action
queued	Due and waiting for an execution slot	Wait; up to three runs can be active at once
running	The session is being created or is executing	Inspect its tools and output
waiting	Approval or user input is required	Resolve the card or answer the question
completed	The session ended	Verify the deliverable and sources
failed / cancelled	Execution failed or was cancelled	Read the run/session error and correct the path, credential, or prompt
configuration-error	A model, workspace, Skill, or MCP was unavailable at runtime	Restore the resource, then run again or wait for the next schedule

✓

Expected resultA repeatable task definition and an inspectable run chain: schedule time, fresh session, approval/tool activity, terminal status, and a verifiable output.

Troubleshooting

• Nothing starts: Confirm the task is enabled, Wordless is still open, and the current time is inside the active range. Use Run now to validate the configuration.
• configuration-error: Re-enable the model, Skill, or MCP; confirm the workspace still exists; then edit and save the automation.
• The run stays waiting: It needs an approval or clarification. Tighten the prompt, or change the approval mode only after a deliberate review.
• A file was overwritten: Name the output path, overwrite policy, and backup rule explicitly. Keep manual approval for valuable material.