VideoMemory

Settings

Configure API keys and model settings. Settings are saved to the local database and persist across restarts.

AI Model Providers

API keys for the AI services that power video analysis.

Required for Gemini models. Get one from Google AI Studio
Required for GPT models. Get one from OpenAI Platform
For accessing models via OpenRouter. Get one from OpenRouter
For Claude models. Get one from Anthropic Console
⚠ API key changes apply immediately to running and new tasks.

Video Processing

Configure the AI model used for analyzing camera streams.

The model used for real-time video analysis. Different models have different speed/accuracy trade-offs.
Only used when Local vLLM model is selected. Default: http://localhost:8100 (set by start.sh). Override if vLLM runs elsewhere.
Stores the captured frame alongside each task note so it can be shown later on the task page. Turn this off to reduce disk usage.
On
Stores a short preroll evidence clip ending on the triggering frame for each task note. Turn this on when you want shareable motion clips, and off to keep CPU and disk usage lower.
Off
⚠ Model and note media storage changes apply immediately to running and new tasks.

External Agent Integration

Configure webhook delivery to an external agent/gateway (for example OpenClaw).

Endpoint for task update events, e.g. http://localhost:18789/hooks/videomemory-alert
Optional bearer token sent as Authorization header.
HTTP timeout for webhook calls. Default is 3.
Suppress duplicate webhook events within this window. Default is 30.
Minimum gap between webhook deliveries. Default is 0.
⚠ Webhook settings apply immediately.
Integration Agreement (How data flows)
  • OpenClaw -> VideoMemory: Fetch the skill at /openclaw/skill.md, then call VideoMemory over HTTP (/api/*) with curl.
  • Skill endpoint from host: http://localhost:5050/openclaw/skill.md.
  • Skill endpoint from a Dockerized OpenClaw: http://host.docker.internal:5050/openclaw/skill.md.
  • Core API base: http://localhost:5050 (or your configured PORT).
  • Typical agent calls: GET /api/devices, POST /api/tasks, GET /api/task/{id}.
  • VideoMemory -> Agent: On detections, VideoMemory POSTs task_update JSON to the configured webhook URL.
  • Auth: If token is set, callback includes Authorization: Bearer ....
  • Tuning: Use timeout, dedupe TTL, and min interval fields above to control callback behavior.

OpenClaw Test Stack

Optional helpers for the bundled VideoMemory + OpenClaw test stack. These values are stored locally for convenience.

Where the OpenClaw UI/API is running (default: http://localhost:18789).
BOT_ID used for VideoMemory alerts and Telegram binding in the test stack.
Stored locally for the bundled OpenClaw stack.
Stored locally for the bundled OpenClaw stack.
Stored locally for the bundled OpenClaw stack.
Stored locally for the bundled OpenClaw stack.
⚠ These values are persisted here even if OpenClaw is offline.