07 · Automations on a schedule

What an automation is

With an automation configured, WorkBuddy runs recurring, repetitive work on a schedule — a daily briefing, a weekly report roll-up, a periodic data tidy — and saves the result to a directory you name.

Mechanically: the automation stores its configuration on the local client (name, prompt, schedule, working directory, state), and at the appointed time starts an agent task as your currently signed-in identity, executing the prompt to query, summarise or process files.

Tasks fire only when the schedule matches, subject to frequency, maximum runtime and system concurrency limits.

Creating one

Open the Automation (自动化) page to see scheduled tasks and execution history. Click Add automation (添加自动化) and fill in:

Field	Description
Name	Distinguishes automations
Prompt	The goal and output requirements
Workspace	Execution directory and where files are saved; defaults to an automation-xxxx space
Permission mode	Default permissions or full access
Model and Skills	Which model, Skills, experts and connectors to use
Schedule rule	Frequency and effective date range
Push to mini program / WeCom	Syncs files to the cloud over a secure channel for delivery

Automations created from an assistant session do not bind the assistant directory

They land in your automation working directory instead.

The design keeps the assistant directory from filling with scheduled output — that directory is for remote conversation, not for accumulating artifacts.

Schedule rules

• Multiple days per week: Monday, Wednesday and Friday at 09:00 — one rule expresses that, no separate task per day
• Multiple days per month: the 1st, 15th and 28th at 09:00

Do not split one job into several automations

The official guidance: combine multiple execution days into one rule rather than creating multiple automations.

The downside of splitting is practical: changing the prompt means changing it in three places, and the number of automated tasks is capped by plan (3 on free, 15 standard, 30 advanced, 99 ultimate). Splitting exhausts that fast.

Start from a template

If you would rather not write a prompt from scratch, use a task template (任务模板). They cover news pushes, weekly report generation, health check-ups, study plans and other common scenarios — pick one and adjust.

Push channels

Once it runs, you need to know the result. Three routes:

To the WorkBuddy mini program

Turn the switch on and results are sent to the mini program when the task completes, readable on your phone.

To a WeCom bot

Enable Push to WeCom bot (推送到企业微信 bot) in the task detail page, below the mini program switch. A dialog offers two sources:

Source	Description	Suits
The assistant's WeCom bot	Reuses the WeCom assistant integration channel	Users who already bound the WeCom assistant
A dedicated automation bot	A separate WeCom bot registered for automation notifications	Notification-only use

Behavioural details worth knowing:

• The current source is shown beneath the switch and can be expanded to change. Unbound sources show a "go bind" entry point and return you to the task afterwards
• With the assistant bot, a WebSocket long connection is the recommended binding method — URL callbacks have an expiry risk and can lapse if not triggered for a long time (the UI flags this with a ⚠ warning)
• If the channel in use is unbound, the source label shows a red dot and a "channel disconnected · rebind" prompt, and the push switch is not turned off automatically
• If credentials are wrong or WeCom is unavailable, a failed notification does not block the task — it still runs, you just do not hear about it

One-way notification

Using the assistant's WeCom bot, the result is sent as a notification only and is not written into the assistant's context — the assistant has no awareness of what ran.

So you cannot reply in WeCom with "add another section to that briefing"; the assistant does not know what you mean. Change the automation configuration in the client instead.

Tasks created through a WeCom bot default to push-on

An automation created in a WeCom bot channel has push enabled and bound to that bot automatically — usable the moment it is created. Tasks created elsewhere default to off.

A full example: the daily briefing

The classic case in the official material, and the fastest automation to show value.

Step one, run it by hand.

Generate today's briefing covering local weather, AI industry developments,
key news and my task summary, and send it to my QQ Mail.

Execution typically covers searching, organising, generating and sending, and takes 2–3 minutes.

Step two, once you have received it, turn it into a schedule.

Create an automation that runs at 08:30 every morning to generate the briefing
and send it to QQ Mail — weather, industry news and a task summary.

Step three, adjust in plain language.

Change the briefing to weekdays only. Keep financial news and AI news, drop
the weather, and make the tone concise and professional.

Always get a manual run working before scheduling it

The official docs stress this, and the reason is simple: automations are unattended. If the delivery path is broken, you find out on day three that you have had no briefings — and the credits for those three runs are already spent.

The full case is A daily briefing that sends itself.

Unattended safety

It runs when you are not there

The official safety note: set prompts involving file writes, deletion or financial operations carefully.

The difference from a manual task is that you are absent when it goes wrong. A "clean up files older than 30 days" schedule with a slightly wrong understanding of the directory will empty that directory while you sleep.

Three disciplines:

One, run new automations at low frequency first. Confirm the agent's understanding and execution match your expectation before increasing the cadence. For a job that should run daily, set it weekly for two weeks first.

Two, keep a traceable record. Models can misunderstand or behave unexpectedly. For anything involving file changes, data changes or outbound sending, keep records you can audit.

Three, third-party services are provided independently. Their stability, billing and results are governed by that provider's status and terms.

Credits

Depends on complexity:

• Simple reminders are cheap
• Code inspections, document roll-ups, connector analysis can consume a lot of tokens

Work out the monthly cost before enabling

A daily task runs 30 times a month. At 30 credits a run that is 900 credits — nearly a quarter of the standard plan's monthly allowance.

Three such automations and your allowance is essentially consumed. Run it manually once and check the real cost before scheduling it.

What is collected

Type	Purpose
Account identity	Attributes the task; the trigger uses your current sign-in to call models, tools, MCP and connectors
Your input	Stores the name, prompt and schedule; the prompt is the execution instruction at trigger time
Usage logs	Creation, start/stop, triggers, results and duration, for status display and audit
Third-party credentials	Only when the task calls MCP, connectors or OAuth services with credentials you already authorised; automation stores no credentials of its own
Local file read/write	Only within the working directory you specified

Data is shared when an automation calls external models, MCP, connectors, OAuth platforms or notification channels.

What should be automated

Three tests, and it must pass all three:

1. Genuinely repetitive — you do it every day or week
2. Rules are unambiguous — steps and criteria are fixed, no judgement calls in the moment
3. No real-time human input needed — nobody has to decide anything mid-run

Good examples:

• A daily work summary at 09:00
• A weekly data roll-up and report
• Regular industry news collection
• Friday at 17:00, summarise this week's commits by module into a weekly report

The counter-example: anything where the answer to "should this go out this week?" varies. A "email the weekly report to my boss every Friday" automation breaks the week the project goes sideways and you need to change the wording. That one should generate locally and let you look before sending.

FAQ

How many automations can I have?

By plan: free 3 (99 during the promo), standard 15 (99 promo), advanced 30 (99 promo), ultimate 99.

Do automations run when my computer is off?

No. Configuration is stored on the local client and the task runs on your machine. The computer must be on with WorkBuddy running.

Can teammates see automations I create in a project?

No. Automations are personal — visible and manageable only by their creator, running in that person's session. To get results to the team, use a push channel.

Next

08 Assistant: drive your PC by phone — take the trigger out of the office.