v2026.04.0

• feature: ai_task.generate_data now accepts image attachments, so you can send camera snapshots or media to vision-capable models and have them generate text based on image content.
• fix: structured outputs on ai_task.generate_data would error if you specified selectors. Now the schema converter handles all selector types correctly.

v2026.03.0

Full Changelog: v0.3.1...v2026.03.0

v0.3.1

• Switch default AI task model from Kimi K2.5 to GPT-OSS-20B

v0.3.0

What's Changed

• Usage monitoring sensors and gateway cost tracking

v0.2.2

Full Changelog: v0.2.1...v0.2.2

v0.2.1

Full Changelog: v0.2.0...v0.2.1

v0.2.0

Breaking changes

• Subentry types changed from single model type to per-platform types (conversation, ai_task_data, ai_task_image). You must remove and re-add the integration after updating.

What's new

• Separate "Add" buttons for each entity type: conversation agent, AI data task, AI image task
• Default conversation and AI task subentries auto-created on setup
• Default provider changed to Workers AI with sensible model defaults:
  • Conversation: GLM-4.7-Flash (fast, tool calling support)
  • AI data task: Kimi K2.5 (high quality structured output)
• Matches the subentry pattern used by Anthropic and OpenAI HA integrations

Full Changelog: v0.1.4...v0.2.0

v0.1.4

First release.

• Conversation agents for 15+ providers via Cloudflare AI Gateway
• AI Task support for non-conversational LLM work
• Streaming responses with native tool calling
• Configurable system prompts, temperature, and top_p per model
• Optional response caching with TTL
• Universal auth -- provider keys stored in Cloudflare, not HA

Requires Home Assistant 2025.7.0+.