Skip to main content
PromptQuorumBuilt for humans. Structured for AI.
Home/Power Local LLM
Voice, Speech & Multimodal

LLM Vision Review: Camera Analysis for Home Assistant With Vision LLMs

·6 min readBy Hans Kuepper · Founder of PromptQuorum · Discovery engine for open-weight & open-source AI

LLM Vision (version 1.7.2, released on 3 September 2026) is a free Home Assistant integration under the Apache-2.0 licence that lets a vision-capable language model describe camera images, videos, live streams and Frigate events. It ships without a model: the analysis runs in the provider you configure. Every statement here comes from the maker's site, its GitHub repository and its privacy file.

Key Takeaways

  • Maintenance: version 1.7.2 was released on 3 September 2026 and the repository was last updated on 17 September 2026 with a beta blueprint for 1.8.0, so no stale-project warning applies.
  • Cost: the integration is free; cloud providers charge per query.
  • AI: no model is bundled, you connect a provider of your choice.
  • Privacy: the privacy file states that the integration contains no telemetry; where frames go depends on your provider, see Data Flow and Privacy.
  • Maker: an individual developer, Valentin Fröhlich, who says the project is not endorsed by or associated with any of the listed providers.

📍 In One Sentence

LLM Vision is a Home Assistant integration that lets a vision LLM describe camera images, video and Frigate events, with the model supplied by a local or cloud provider you choose.

💬 In Plain Terms

Your camera or Frigate hands a picture to an AI model that you set up separately, and the answer comes back as text that Home Assistant can show, store or act on.

📌Note: This review is based on llmvision.org, the GitHub repository and its privacy file, checked on 1 November 2026. PromptQuorum has not installed, run or benchmarked the integration, so speed, detection accuracy and description quality are not assessed.

Read in:
🇺🇸en🇩🇪de🇫🇷fr🇯🇵ja🇨🇳zh🇪🇸es🇧🇷pt🇸🇦ar🇰🇷ko

LLM Vision is a free Home Assistant integration that sends camera images, video clips and Frigate events to a multimodal language model and returns a text description that notifications, automations and an event timeline can use. It does no AI work itself: you connect a provider, either Ollama, Open WebUI or LocalAI on your own hardware or a cloud service. The maker is Valentin Fröhlich. This review is based on the website, the GitHub repository with its releases and privacy file, and the site's page on the Glimpse model, checked on 1 November 2026; PromptQuorum has not tested the integration hands-on.

LLM Vision Review: Camera Analysis for Home Assistant With Vision LLMs

What Is LLM Vision?

LLM Vision is a custom integration for Home Assistant that passes images from cameras, image entities, video files and Frigate events to a multimodal language model. The site calls it "Visual Intelligence for your Home", and the GitHub repository was created on 14 May 2024.

It is an add-on for an existing smart-home setup, not a standalone app: you need a running Home Assistant, the HACS community store and an AI provider. The maker is Valentin Fröhlich, credited in the site's copyright line; the repository lists a second code owner on GitHub.

Get It

LLM Vision installs from HACS inside Home Assistant; there is no separate download, store app or installer.

Option
How to get it
InstallLLM Vision on GitHub, added through HACS
SetupAdd it under Devices & services, set up the /media folder for snapshots, then add an AI provider entry
Local providersOllama, Open WebUI or LocalAI on your own hardware
Cloud providersOpenAI, Anthropic, Google, OpenRouter, AWS, Groq or Azure AI, billed per query by the provider
PriceThe integration is free; donations via Buy Me a Coffee are optional
DocumentationGetting started guide
Websitellmvision.org
Privacy policyPRIVACY.md on GitHub
Source codeApache-2.0 licence on GitHub

Version reviewed: 1.7.2 (3 September 2026, GitHub). The integration manifest in the repository still says 1.7.1 and the website says v1.7.0, so both lag the release. No minimum Home Assistant version is stated in the repository, and Home Assistant Container users may need to mount the /media folder.

How to Get Started

The steps follow the maker's README; PromptQuorum has not run them.

  1. 1
    Install the integration
    Why it matters: Install LLM Vision from HACS and restart Home Assistant.
  2. 2
    Add it to Home Assistant
    Why it matters: Search for LLM Vision under Devices & services and submit with the default settings.
  3. 3
    Prepare the media folder
    Why it matters: Snapshots are stored in /media; Home Assistant Container users may need to mount it.
  4. 4
    Add an AI provider
    Why it matters: Return to the integration page and add an entry for your provider, for example Ollama on your own server or a cloud API key.

Features Confirmed by the Sources

Every item below comes from the maker's site or repository; none has been independently tested.

  • Four actions. Images analyses cameras, image entities or files; Videos extracts frames from a video file or Frigate event; Streams captures frames from camera entities; Data analyses an image and updates a number, text, select or boolean helper.
  • Notifications. Camera feeds and Frigate events are analysed in near real time, and results can feed notifications and automations.
  • Timeline and memory. Events are stored only if "remember" is on; with "important" they are classified and kept only if judged important, and appear as calendar entities. Remembering events needs a conversation agent such as Extended OpenAI Conversation.
  • Image pipeline. The site says it picks relevant frames and optimises them for inference, and a downscaler reduces resolution and token use.
  • Glimpse-v1. The maker's own 4-billion-parameter open vision-language model for summarising home-security events, in beta, on Hugging Face and the Ollama Library (ollama run llmvision/glimpse-v1).
  • Languages. English, German, Dutch, French, Spanish, Portuguese, Italian, Polish and Swedish, with more added over time.

Data Flow and Privacy

The integration sends images only to the provider you configure, so where they end up depends entirely on that choice. With Ollama, Open WebUI or LocalAI on your own hardware the analysis stays with you; with OpenAI, Anthropic, Google, OpenRouter, AWS, Groq or Azure AI the frames travel to that company's servers and fall under its terms.

The privacy file (effective 15 February 2026) says the Home Assistant integration contains no telemetry: the maker receives no usage data, cannot see which providers you configure and has no access to your Home Assistant data. No accounts are needed, and only the website uses Google Analytics. The site adds that analysed events never leave your Home Assistant instance, while the integration manifest classifies it as cloud polling.

📌Note: These are the maker's declarations, not audit results, and PromptQuorum has not inspected network traffic. The site's statement about events refers to stored events, not to the calls made to a provider.

Who Should Use It

  • Home Assistant users with cameras who want a text description of a doorbell or driveway event in the notification.
  • Frigate users who want events summarised and searchable in a timeline.
  • Ollama, Open WebUI or LocalAI operators who want frames analysed on their own hardware.

What We Could Not Verify

  • Performance. Detection accuracy, latency, hardware needs and the cost per query are untested, and the site names no hardware requirement.
  • Fully local operation. Neither the site nor the README spells out an end-to-end local setup; they list Ollama, Open WebUI and LocalAI as providers and recommend a local provider for the most private setup.
  • Home Assistant version. No minimum version is stated in the repository.
  • Glimpse-v1 licence. The site does not name a licence for the model and gives no hardware figures.

Competitors and Alternatives

Home Assistant Assist

Platforms:
Linux, web, iPhone, Android
Price / license:
Free / Apache-2.0
Key difference:
Built-in voice assistant; LLM agents through Ollama or OpenAI
Articles about Home Assistant Assist (1)

Confirm each app's current price, license and platforms on its own page before choosing.

Compare in directory

Frequently Asked Questions

Can I use a provider that is not on the list?

The site says providers with an OpenAI-compatible API endpoint can also be used, possibly with additional setup.

Is there a paid version or subscription?

None is listed; the site shows only an optional Buy Me a Coffee donation link, and the integration itself is free.

Where do I get help or report a bug?

The site links GitHub Discussions, a Discord server and a Home Assistant community thread, and lists hello@llmvision.org for contact. Its "Report a bug" link points to the separate dashboard-card repository, not to the main one.

Verdict

LLM Vision turns Home Assistant camera images, video and Frigate events into text through a vision model you choose, it is free under Apache-2.0, and its repository was updated on 17 September 2026. Against that, it brings no model of its own, cloud providers receive your frames, no minimum Home Assistant version or hardware requirement is stated, and accuracy and speed are untested. It suits Home Assistant owners who can run Ollama, Open WebUI or LocalAI themselves; anyone sending frames to a cloud provider should read that provider's terms first.

Sources

  • LLM Vision website — features, providers, privacy and install notes, checked 1 November 2026.
  • Glimpse-v1 page — model description and availability, checked 1 November 2026.
  • LLM Vision on GitHub — README, licence, manifest and commits, checked 1 November 2026.
  • GitHub release v1.7.2 — release of 3 September 2026, checked 1 November 2026.
  • Privacy policy — telemetry and data handling, effective 15 February 2026, checked 1 November 2026.

← Back to Power Local LLM

MCP