Top 200 ReposEXPLORER
Browse agents, categories and pages
Back to results
Planned, not doneThe source describes an intention, not a finished result.

Get written summaries and threat levels for camera activity with Frigate’s GenAI review

Frigate can send the frames of each finished alert to a vision model and save a short title, a scene description and a threat level of 0 to 2, which then appear in its notifications and review screens.

Try the promptRead the original (opens in a new tab)
Upvote 0. Sign in to upvoteSign in to upvote · One vote per signed-in person. Site votes are separate from GitHub stars.

The problem

How it works

Frigate first records and detects objects on its own; the GenAI review step comes after. You configure a vision-capable provider, such as a local Ollama or llama.cpp server or a hosted API, then turn on review summaries globally or per camera. When an alert ends, Frigate sends up to 20 frames and gets back JSON with a title, a scene narrative, a two-sentence short summary, a confidence score, any of your own concerns it noticed and a threat level. You can rewrite the activity context that defines normal, suspicious and critical behaviour for your property, add concerns such as animals in the garden, choose the summary language, and ask for a report of everything marked suspicious over a period, for example while you were away.

Results and limitations

This is Frigate’s own documentation with no measured results, so it is marked planned.

Before you try it

Frigate warns that running these models on a CPU is impractically slow, and its docs recommend a GPU and a vision model such as qwen3-vl. Recording frames give 480p images instead of 180p previews but use more tokens. A hosted provider receives frames from your cameras, so a local model keeps footage at home. Keep the activity context general, because an overly specific prompt sways the result, and treat the threat level as a prompt to look, not a verdict. Results depend on your setup and were not independently reproduced by Top 200 Repos.

Tools and repositories

Tools
Frigate · Ollama, llama.cpp or a hosted vision model
Prerequisites
A running Frigate install with alerts, and a vision-capable model, ideally on a GPU.

Try it yourself: Write an activity context for your cameras

Fill the fields to adapt the prompt. Empty fields stay as visible [placeholders].

Preview
Help me write an activity context for Frigate’s GenAI review summaries. My property is [Your property]. Normal activity includes [Normal activity]. I want to be told about [What to flag]. Write it in three short sections: normal (level 0), suspicious (level 1) and critical (level 2), with general rules rather than descriptions of specific people, and keep it under 200 words.

Paste the result into Frigate’s activity context prompt and adjust it after a week of real alerts.

Related videos

About ollama/ollama

About ollama/ollama

Independent creators. Listing a video implies no endorsement, sponsorship or partnership.

Sources

  1. Review Summaries (opens in a new tab)

    Project documentation · docs.frigate.video · Undated · Source checked 12 Nov 2026, 18:05 UTC

    What this source supports
    • Review summaries are generated for alert review items when activity ends and shown in Frigate’s notifications and UI
    • Each summary is JSON with title, scene, shortSummary, confidence, other_concerns and potential_threat_level 0, 1 or 2
    • The activity context prompt can be set globally or per camera and should not be overly specific
    • Recording frames are extracted at 480px versus preview frames, capped at 20 frames; additional concerns and preferred language are configurable
    • Reports summarise suspicious review items over a period, through the API or the frigate.review_summarize service in Home Assistant
  2. Configuring Generative AI (opens in a new tab)

    Project documentation · docs.frigate.video · Undated · Source checked 12 Nov 2026, 18:06 UTC

    What this source supports
    • Providers include Ollama, llama.cpp and OpenAI-compatible APIs, and a vision-capable model is required
    • Running Generative AI models on CPU is not recommended because of high inference times
    • qwen3-vl is among the recommended local vision models
Report a problem

Planned, not done

Turn a camera snapshot into a sensor with Home Assistant’s AI Task

Home Assistant 2025.8 added AI Task: an automation sends a camera image to your chosen AI with a question, gets a structured answer back and stores it, for example as a count of birds in a coop.

home-assistant.iohome-assistant/core, ollama/ollama
Planned, not done

Auto-tag receipts and bills in Paperless-ngx with paperless-ai

paperless-ai suggests titles, tags, correspondents and dates for documents in a Paperless-ngx archive, using a local Ollama model or a cloud LLM, with manual review and undo.

github.comollama/ollama
Reported by source

Extract invoice fields locally with a vision model on Ollama

Stephen Sebastian replaced a paid cloud OCR API with Gemma 4 E4B on Ollama to pull date, amount and description from contractor invoice photos on his laptop.

dev.toollama/ollama
Planned, not done

Record and summarise meetings on your own machine with Meetily

Meetily records calls, transcribes them on the device with Whisper or Parakeet, and writes summaries with a local Ollama model or a cloud model you pick.

github.comollama/ollama
Reported by source

Run an open-weight model locally and call it from your code

Install Ollama, pull an open model and call it through the local REST API so prompts stay on your machine.

github.comollama/ollama
Planned, not done

Run OpenAI’s Codex CLI on open models served by Ollama

Ollama’s docs show how to point Codex CLI at local tool-capable models such as gpt-oss with one launch command or a saved profile, switch models inside a session and turn off web search.

docs.ollama.comopenai/codex, ollama/ollama