Home automation
Home automation's oldest complaint: why did the heating turn on at 3am? Record every automation decision as a memory and the answer is one query away — with the decision's own evidence cited.
Capture decisions from your automations
From Home Assistant (a rest_command), Node-RED, or a plain shell hook:
# Home Assistant configuration.yaml
rest_command:
memory_capture:
url: "http://127.0.0.1:4040/v1/capture/task"
method: POST
headers:
Authorization: "Bearer YOUR_API_TOKEN"
content_type: "application/json"
payload: >
{"project": "house", "task_title": "Automation decision",
"user_prompt": "Record an automation decision.", "writer_id": "home-assistant",
"agent_summary": "{{ summary }}",
"structured_candidates": [{
"canonical_text": "{{ detail }}",
"summary": "{{ summary }}",
"memory_type": "incident", "confidence": 0.9, "importance": 3,
"tags": ["automation", "{{ area }}"],
"sources": [{"source_kind": "note", "excerpt": "{{ trigger_desc }}"}]}]}Call it from an automation's action with the decision and its trigger:
- service: rest_command.memory_capture
data:
summary: "Heating on in the studio at 03:12"
detail: "Heating turned on in the studio at 03:12 because the temperature dropped to 14.5°C, below the 15°C frost-guard threshold, while the away-mode override was inactive."
area: "studio"
trigger_desc: "sensor.studio_temperature = 14.5"Run curl .../v1/curate -d '{"project": "house"}' on a schedule (or after each capture) to promote the captures.
Ask the house
memory query --project house --question "Why did the heating turn on in the studio last night?"The answer cites the exact decision memory, with the trigger reading as provenance. Durable configuration knowledge belongs here too — capture threshold changes and device quirks as decision/environment memories, and six months later "why is the frost guard at 15 degrees?" has a real answer.
Everything stays on your LAN: the service binds localhost by default, works keyless, and runs happily on a Raspberry Pi–class machine next to your automation hub.