Sound Event Memory

Indexes and recalls non-speech ambient audio events (glass break, alarms, dog barks) across devices with timestamps and confidence scores for incident review.

Sound Event Memory — BlueColumn Skill

Video tells you what a camera saw. Sound tells you what actually happened. This skill indexes non-speech audio events from edge devices (glass break, alarms, dog barking, raised voices, doors, gunshots) into persistent per-device memory, then recalls them by question instead of by scrubbing footage.

Buyer: property managers, alarm/monitoring companies, short-term rental operators. Why they pay: incident review today means hours of footage; there is no searchable record of what a site sounded like. Sound-event memory gives timestamped, citable evidence. Pilot: $499 one-time, 14 days, up to 10 devices.

Setup

Get a key at https://bluecolumn.ai/login (free tier, no card). Keys look like bc_live_*. Canonical API base: https://api.bluecolumn.ai.

Index an audio window

Send the audio; BlueColumn transcribes speech and detects non-speech events. Pass the signals you care about so the detector focuses.

bash

curl -X POST https://api.bluecolumn.ai/remember \
  -H "Authorization: Bearer $BLUECOLUMN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "audio_url": "https://example.com/site7/front-door-2026-09-30-0213.wav",
    "text": "SOUND EVENTS — site=site7 device=front-door, window 02:13-02:19 MST. Events: glass_break (02:14:07, conf 0.91); raised_voices (02:14:20-02:15:02, 2 speakers); door_force (02:14:55, conf 0.84). No speech intelligible. Handled by: none (flagged).",
    "title": "site7 front-door 2026-09-30 0213 sound events"
  }'

Recall across devices and time

Ask a question, not a query string. The answer cites the device and timestamp.

bash

curl -X POST https://api.bluecolumn.ai/recall \
  -H "Authorization: Bearer $BLUECOLUMN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"q": "Every glass-break or forced-door event at site7 in the last 30 days, with timestamp and confidence."}'

For live streaming devices, ingest chunks with /streaming-audio (see the streaming-audio-memory skill) and this memory indexes events as they arrive.

Log an incident decision

Keep the human call next to the evidence so audits are complete.

bash

curl -X POST https://api.bluecolumn.ai/note \
  -H "Authorization: Bearer $BLUECOLUMN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"text": "site7 front-door 02:14 glass-break: confirmed false alarm (mirror fell in wind). Dispatched no unit. Reviewed by ops on 09-30.", "tags": ["incident", "site7", "false-alarm"]}'

Workflow

1. Capture — stream or upload ambient audio per device; store the detected events with timestamps and confidence.
2. Flag — mark events that need a human (glass break, alarm, gunshot) versus reviews.
3. Recall — answer incident questions across all devices and dates in plain English.
4. Prove — cite the exact device and timestamp for insurance, police, or tenant disputes.

Billing

If the API returns 402 or 429, read topup_url / upgrade_url from the error body and surface that link to the user — do not retry.

Docs

API reference: https://bluecolumn.ai/docs — request fields are text, q, tags (not content/query/note).