Tutorial: from zero to a self-operating capability

A complete walkthrough — connect PowerExecutor to PowerMarketing, watch it in shadow, approve its first real action, then let one capability run on its own. About 20 minutes.

Part 1 — Connect & configure

1. Connect the console

   Open the console, go to Settings, set the API URL, Save & connect, and wait for ● Connected. Pick powermarketing in the left-pane Product selector.

2. Set the AI model

   Configuration → AI: set Provider + Model + API key + API base URL. Example for DeepSeek:

   Field	Value
   Provider	deepseek
   Model	deepseek-chat
   API base URL	https://api.deepseek.com
   API key	your key

   Save. (First save asks for the Supervisor PIN.)

3. Set the product connection

   Configuration → Connection — powermarketing: Username admin, Password (the real PM admin password), Brand id (e.g. 1, or blank for the first brand). Save.

Part 2 — Observe in shadow (safe)

1. Run a cycle

   Dashboard → ▶ Run now (on the powermarketing card). Every capability is at 👁 Watch only, so nothing on PowerMarketing changes — PowerExecutor only reads signals and reasons.

2. Read the report

   Latest report shows: the signals it read, the situations the AI concluded, and — for each proposed action — that it was Simulated (no real change). If the Needs attention card is empty, you're connected and reasoning correctly.

   If attention shows "Cannot connect to the product" (401) → the PM password/brand is wrong. If it shows "The AI model call failed" → the Provider is empty or the model/key is wrong. Fix in Configuration and re-run.

3. Repeat for a few cycles

   Let it run several times (or wait for the daily cycle). Read each report. You are learning to trust its reasoning before you give it any real power.

Part 3 — Approve its first real action

1. Promote one capability to Ask-me-first

   Open Permissions. Pick a low-risk, reversible capability (e.g. update_post), set it to 🙋 Ask me first, click Set.

2. Run a cycle

   Dashboard → ▶ Run now. Now, when the AI proposes that capability, it doesn't simulate — it puts the action in the approval queue.

3. Review and approve

   Open Approvals. Each row shows the capability, the situation that triggered it, and the exact parameters. Read it. If it's sound, click Approve — PowerExecutor executes it for real, verifies it, and writes a receipt. Otherwise Reject.

   Irreversible capabilities (e.g. send_broadcast) always require approval, at every level — by design.

Part 4 — Let it run on its own

1. Promote to Act-alone

   Once a capability has proven itself across several approvals, set it to 🤖 Act alone in Permissions. From now on it acts without asking.

   PowerExecutor also promotes automatically after three consecutive clean outcomes, and demotes on any failure — so trust is earned, not assumed.

2. Enable the daily schedule

   Configuration → Operations: confirm Enabled products includes powermarketing and set Cycle hour (UTC). The server now runs a full cycle once per day automatically.

3. Supervise, don't operate

   Your job is now: skim the daily report, clear the Approvals queue, and step in with the kill switch if anything looks wrong. PowerExecutor does the rest.