Quick start

1. Run the API

Locally:

cd powerexecutor-api
python -m uvicorn app.main:app --port 8040

Set the reasoning model and (optionally) a supervisor PIN via environment variables:

LLM_MODEL=anthropic/claude-sonnet-4-6   # any LiteLLM model id, local ones too
ANTHROPIC_API_KEY=...                    # key for your chosen provider
POWEREXECUTOR_PIN=1234                    # optional; gates consequential actions

In production the API runs at https://executor-api.planningpowertools.com and the console at https://planningpowertools.com/executor/.

2. Open the console & connect

Open http://localhost:8040/ui/. On first run the Settings view opens. Fill in:

• API URL — where the API runs (http://localhost:8040 locally).
• Product — powermarketing.
• Supervisor PIN — the shared secret set on the server as POWEREXECUTOR_PIN; required for approvals, autonomy changes, Configuration saves and the kill switch. Details.

Click Save & connect. The indicator in the sidebar turns green (● Connected).

3. Review autonomy levels

Open Autonomy. Every capability starts at Shadow — safe, since Shadow never touches the product. Leave them there for your first runs.

4. Run a cycle

On Overview, click ▶ Run cycle now. PowerExecutor observes the product, reasons about it, and — at Shadow level — records what it would have done.

5. Read the report

Open Latest report to see the situations it identified, the actions it considered, and any rejected reasoning or unknowns. This is how you learn what the operator is thinking before you grant it more freedom.