Autopilot

Scheduled tasks

Set the cadence and the agent runs itself — reports, audits, sourcing, with no one at the desk.

What it does

Hand repetitive operations to a schedule: set the cadence in the UI and the system runs on time. This is what "24/7 operations that never quit" looks like in practice — SEO weekly reports, AEO audits, keyword and LinkedIn sourcing can all be set to run daily, weekly or on a fixed interval, working whether or not you are at the screen.

Two task types with clean safety boundaries: an "action" is self-contained analysis and generation that runs to completion on its own and only produces, never sends; a "command" hands one line of natural language to the agent, but any tool that writes or sends externally lands in the confirmation queue for your approval. Scheduled does not mean unsupervised.

The most useful built-in action is "write a blog every day": on schedule it pulls industry news, has the AI pick a hot topic, drafts to an outline, adds a cover image and saves a site draft for your review — drafts only, unless you explicitly set direct publish. You can even say "write me a blog draft daily" in chat and the agent configures the template and schedule itself (that creation step still passes confirmation). Want the agent to run a whole task unattended? Each task has its own autonomous-execution switch — but high-risk actions that touch the live site or spend money, like publishing an article, uploading media or generating an image, always force human confirmation no matter the setting, and even credit-burning actions carry a rolling 24-hour cap so a misconfigured interval can't drain your budget.

Core capabilities

Runs on time, on its own

Set the cadence once; reports, audits, sourcing and a daily blog fire daily, weekly or on interval, even when you are away.

A blog draft every day

News → topic → draft → cover image → saved draft, hands-off; drafts only by default, and one sentence lets the agent set it up.

Autonomous run with a hard gate

Let the agent run a task unattended, but publishing, uploads and image generation — anything that spends or touches the live site — always force human confirmation.

Feature checklist

  • Three cadences: fixed interval, daily and weekly; daily and weekly resolve against the operator's local timezone
  • Predefined actions: weekly report, AEO audit, keyword sourcing (from template), LinkedIn sourcing (from template), daily auto-blog (saved as draft)
  • Daily auto-blog: news → AI topic pick → outline → draft → optional cover image → saved site draft; 400-4,000 words, 1,200 by default, drafts only by default
  • Any write or send tool in a command task passes through the policy guard into the confirmation queue — never auto-sent
  • A per-task autonomous-execution switch (off by default): once on, the agent runs multi-step to completion, each run bounded by step, cost and wall-clock guardrails
  • Hard gate on high-risk actions: publishing or editing articles and products, uploading media, schema injection and image generation — eight in all — always queue for confirmation, even unattended
  • Credit-burning actions carry a rolling 24-hour run cap, hot-editable with value clamping, so an over-frequent interval can't drain the budget
  • "Run now" shows live stage progress, and re-triggering a task already running is skipped automatically — no double credit burn
  • A 60-second polling loop advances the next run time before executing, so slow tasks never double-trigger
  • Sourcing is expensive to configure once, so it can be saved as a search template and re-run on schedule, with templates renamed, deleted or inspected by condition
  • The whole scheduler API enforces operator-or-above, with a global on/off switch

Want to see this module run?

Book a demo and we'll run this module on real data.