v1.10.0
What's new in v1.10.0
Optionally summarise & categorise your pending HACS updates with AI (#67).
Draft for testing — not offered to users until published.
Added
- AI summarise & categorise updates — opt-in via Settings → Options: enable it and pick either an AI Task entity (HA 2025.7+) or a Conversation agent. Each pending update is tagged with categories (Bug fixes, New features, Documentation, Translations, Breaking changes, Dependencies, Other) and gets a one-line summary.
- Update Action Tracker card: category badges per update, a category filter, a sort-by-category toggle, and the summary in the expanded row.
Compatibility
- Works with two AI source types — the integration auto-detects which you picked:
- AI Task entity: requests structured output, falling back to plain-text + JSON parsing if the backend can't do strict schema.
- Conversation agent: a single
conversation.processcall, parsing JSON from the reply. This often works for self-hosted models where the AI Task path doesn't.
- Loose category labels are normalised onto the fixed set. Results are cached per component+version (the hourly scan only calls the AI for new/changed updates). Failed updates aren't cached, so flaky local models recover on later scans.
- If categorisation fails, the scan still completes, those updates show no categories, and a warning explains why.
Tested with
- ✅ qwen3.5-4b-vision (AI Task, LocalAI) — clean, exact category tags.
- ✅ Azure OpenAI GPT-5.x (AI Task) — works; loose tags normalised.
- ✅ gemma-4-e4b-it via a Conversation agent — categorises correctly. (Via the AI Task path it fails inside the third-party extended_openai_conversation bridge — use the Conversation agent for such models.)
- Recommended: official OpenAI / Ollama / Anthropic / Google Generative AI providers give the most reliable results; if one source type doesn't work for your model, try the other.
Upgrade notes
- Off by default — unchanged behaviour until you enable it and choose an AI entity.