# ha-llmvision: one integration, eleven provider backends, and three releases titled Bug Fixes

> A Home Assistant custom component that sends images, video files, live camera feeds and Frigate events to multimodal models and turns the answers into sensors and a timeline. Installation is seven steps through HACS, the seventh of which is where you add a provider credential, and a pytest suite sits next to a component that is otherwise configuration files.

**valentinfrlch/ha-llmvision** — Visual intelligence for your home.

- Repository: https://github.com/valentinfrlch/ha-llmvision
- Website: https://llmvision.org
- Stars: 1,479 · Forks: 146
- Language: Python
- License: Apache-2.0
- Published: 2026-09-10 · Updated: 2026-09-10 · Language: en
- Canonical page: https://hysenlabs.com/projects/valentinfrlch-ha-llmvision

## Eleven provider backends behind one component

The provider line is the longest list in the file and it is worth reading as a compatibility promise rather than a feature count. It names OpenRouter, OpenAI, Anthropic, Google Gemini, AWS Bedrock, Azure and Groq, then three local runtimes, Ollama, Open WebUI and LocalAI, and closes with any provider that exposes OpenAI compatible endpoints. That last clause does more work than the named entries, because it means a self-hosted server with an OpenAI-shaped API needs no code change. What the component does with the model is the same either way: it answers questions and produces descriptions of images, video files, live camera feeds and Frigate events from a prompt you supply, it remembers people, pets and objects, and it updates sensors based on data extracted from those sources. Frigate is called out separately in the description, so an existing Frigate installation is treated as a first-class input rather than a camera type.

## Setup is seven steps and the seventh is the credential

The quick start is a numbered list, and the order matters because of what each step assumes. The component is available in the default HACS repository, so no custom repository needs adding first. Then: install LLM Vision from HACS, restart Home Assistant, search for LLM Vision in Settings under Devices and services, and press submit to continue with default settings. Step five is the one with a wrinkle: set up the media folder, because the component uses the more secure /media folder for storing snapshots, and if you run Home Assistant Container you may need to mount a folder to /media in your container settings. Step six returns you to the integration page and step seven presses Add Entry to add your first AI provider. So default settings plus a mounted media folder gets you a working integration with no provider configured, and the provider is a separate entry rather than part of the initial form.

## The timeline is the durable part, not the description

The description of a single frame is transient; the timeline is what makes the component worth configuring. LLM Vision keeps a timeline of camera events, which you can display on your dashboard or ask Assist about later, and an optional Timeline Card is offered for that dashboard. A blueprint is provided on top, described as an easy way to get camera event notifications intelligently summarized by AI, and it can store events in the timeline so you can see what happened. The component's own design surfaces a little in the tree: timeline_openapi_specs.yaml sits at the repository root, an interface description for the timeline that the card consumes, which means the timeline is treated as something with a published shape rather than a private blob. The provider and the timeline are separable concerns in the setup too, since the provider is added as its own entry after the integration exists.

## A pytest suite next to a component made of configuration

Home Assistant custom components are conventionally YAML, Python and a manifest, and this repository adds a test layer to that. The tree carries pytest.ini, requirements-test.txt, a run_tests.sh script, a tests/ directory and a README_TESTING.md, alongside the component itself under custom_components/ and the blueprint under blueprints/. A shell entry point for the suite is a small thing that tells you the tests are meant to be run on the command line rather than only in an editor, and a separate testing readme at the root suggests the process has been written down rather than left in somebody's shell history. Also present is a benchmark_visualization/ directory, and translator.py at the root, which is the script shape a Home Assistant translation pipeline usually takes.

## Three releases in six weeks, all titled Bug Fixes

The release record carries no functional information. v1.7.1 is titled Bug Fixes and Performance Improvements and is dated 2026-08-04. v1.7.2-beta.1 is titled Bug fixes, dated 2026-08-30. v1.7.2 is titled Bug Fixes, dated 2026-09-03. Three releases inside a month, two of them sharing the number 1.7.2 with a beta preceding its own final, and not one of the titles naming a change you could look for. For a component that people install through a package manager rather than pin, that is workable, since the version tells you when to update and not what changed. For anyone tracking behaviour changes, or writing a bug report against a version, it means the only record is the issue tracker. The last commit on the default branch is dated 2026-09-17, two weeks after the newest tag.

## Privacy documents in two languages, and a translation script

The repository root is unusually well documented for a custom component. It holds PRIVACY.md and PRIVACY.zh-CN.md, a Chinese project guide, a Chinese README beside the English one, NOTICE, CONTRIBUTING.md and hacs.json, which is the manifest HACS reads to list the integration. The Chinese files sit next to their English counterparts rather than in a locale directory, and translator.py at the root is the script that would produce them. That is a deliberate translation workflow, which matters for a component whose user base is international and whose documentation is mostly prose. It is also the honest place to look for data handling, because the README itself makes no claim about where camera images are sent once a provider is chosen, and the choice is the user's: a hosted model and a local one are both listed in the same sentence.

## Debug logs sit behind a settings toggle

The bug reporting section is short and specific. If you hit a bug and have followed the instructions, file a report, check open issues first, and include debug logs, which can be enabled on the integration's settings page. Feature requests go through a separate path, and both are collected by issue templates at the repository's new-issue chooser. The detail that matters for anyone reproducing a problem is that logging is off until you turn it on, so a report without logs means a second round trip. The support section, for its part, asks for a star and offers a coffee, which tells you where maintenance effort is expected to come from.

## Conclusion

ha-llmvision fits a Home Assistant household that already runs cameras and wants descriptions, object memory and event summaries without building the pipeline itself, and the local provider options mean it does not have to be a cloud arrangement. Two things to settle first. Where the images go is your decision, not the integration's: the same setup can call a hosted model or a local one, and the README says nothing about data handling while the repository carries privacy documents in two languages. And the release record is thin, since three releases in six weeks are all titled Bug Fixes and a beta shares its number with the final it precedes, so version numbers here tell you when something shipped and nothing about what changed. Apache 2.0, default branch main, last commit 2026-09-17.

## FAQ

### What is LLM Vision for Home Assistant?

A Home Assistant integration that uses multimodal large language models to analyze images, video files, live camera feeds and Frigate events. It answers questions and describes them from your prompt, remembers people, pets and objects, and updates sensors based on data extracted from camera streams, images or videos.

### Which AI providers can LLM Vision use?

OpenRouter, OpenAI, Anthropic, Google Gemini, AWS Bedrock, Azure, Groq, Ollama, Open WebUI, LocalAI, and any provider that exposes OpenAI compatible endpoints, so a self-hosted server with an OpenAI-shaped API needs no code change.

### How do I install LLM Vision?

Seven steps from HACS, which is where the integration is already listed. Install LLM Vision, restart Home Assistant, find it under Settings and Devices and services, press submit to continue with default settings, make sure a folder is mounted at /media for snapshots, then return to the integration page and press Add Entry to add your first AI provider.

### Does LLM Vision send my camera images to a cloud service?

That depends on the provider you add. The file lists hosted services such as OpenAI, Anthropic, Gemini, Bedrock, Azure, Groq and OpenRouter alongside local ones such as Ollama, Open WebUI and LocalAI, and it stores snapshots in the /media folder, which it describes as more secure. The README makes no data handling claim, so the privacy documents in the repository are the place to look.

### What is the LLM Vision timeline?

A record of camera events that LLM Vision keeps so you can display them on your dashboard or ask Assist about them, with an optional Timeline Card for that purpose. A blueprint is provided as well, which summarizes camera event notifications with AI and can store those events in the timeline. The tree includes an OpenAPI description of the timeline at its root.

## Sources

- [License: Apache-2.0](https://github.com/valentinfrlch/ha-llmvision/blob/main/LICENSE)
- [Project website](https://llmvision.org)
- [README](https://github.com/valentinfrlch/ha-llmvision/blob/main/README.md)
- [Releases](https://github.com/valentinfrlch/ha-llmvision/releases)
- [valentinfrlch/ha-llmvision on GitHub](https://github.com/valentinfrlch/ha-llmvision)

---

Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/valentinfrlch-ha-llmvision