Skip to content

v3.1.0

Choose a tag to compare

View all tags
@pajeronda pajeronda released this 05 Feb 23:56
3.1.0
f5ee223

Release Notes

Breaking Change

  • SDK requirement bumped to >= 1.6.0 (required for tool_call_id linking and image_url support).

New: HassPhotoAnalysis — Custom tool

Grok can now see through your cameras. When a user asks something like "look at the garden camera" or "is there anyone at the front door?", Grok automatically:

  • Captures a snapshot from the specified camera entity
  • Sends the image to Grok Vision for analysis
  • Returns a natural language description in the conversation

The tool is available automatically when the camera domain is exposed to the assistant. Snapshots are saved to /media/xai_snapshots/ with camera name and timestamp (e.g. garden_20260131_231500.jpg), allowing the user to review or delete them at will.

New: IntentExecution — Your custom intents become tools!

Support in mode tools for custom intents defined in /config/custom_sentences/.

The integration automatically scans your custom sentences and builds a topic list for Grok, so it knows when to route a request through Home Assistant's intent system.

Two discovery strategies (no configuration needed):

  1. call_* slots — If your intents use slot values like call_weather, call_garbage, the topic is extracted automatically (e.g. weather, garbage).
  2. Intent name fallback — If no call_* slots are found, the intent name is used as topic (e.g. SetVolume → set volume, GetWeather → get weather).

Example with call_* slots:

language: "en"
intents:
  TriggerAssistMessage:
    data:
      - sentences:
          - "[<how_is>] [<the>] (weather|forecast) [for | in] {city_request}..."
        slots:
          call_message: "call_weather"

Example without call_* slots (works automatically):

language: "en"
intents:
  SetVolume:
    data:
      - sentences:
          - "(set|change) volume to {volume} [percent]"

The tool activates automatically as soon as you have at least one custom intent.

Updated: Default vision model

The default vision model is now grok-4-1-fast-non-reasoning, replacing the legacy grok-2-vision-1212. Grok 4.x models natively support image analysis without a dedicated vision model, at a lower cost ($0.20/1M tokens vs $2.00/1M).

Updated: Default image generation model

The default image generation model is now grok-imagine-image. This new model offers significantly higher image quality and a lower cost ($0.02 per image).

Improved: Web search location context

When live search is enabled, the user's configured home location is now passed to the xAI web search API for better localized results (city, country).

Bugfixes & Improvements

Various bugfixes and improvements across image services, pricing sensors, and tool call handling.

Note: Pricing discrepancy (Image models)

Some users may see a higher price in the Home Assistant pricing sensors compared to the xAI website (e.g., grok-imagine-image output showing ~$0.033 per image). The integration displays raw API pricing returned by xAI; if the site shows a different value, the mismatch is on the xAI server side.

Notice

xAI will retire grok-2-image-1212, grok-3-mini, and grok-2-vision-1212 starting February 28, 2026.

Assets 2
Loading