⚠️ Required for Home Assistant 2026.8.0+
The Selora AI panel does not render correctly on Home Assistant 2026.8.0 and later without this release. HA no longer always gives the panel container a definite height, so the panel shell resolved to auto and became content-sized: the session list laid out every row at full length, the page grew to several screens tall, and the header scrolled off the top. Installs with a long chat history saw the worst of it.
If you are running HA 2026.8.0 or newer, upgrade to v0.14.0.
The panel shell is now clamped to the viewport (100vh with a 100dvh refinement, plus overflow: hidden). The clamp only applies in the failure case — when HA does size the container, height: 100% already resolves smaller and nothing changes, so panels HA deliberately constrains are unaffected.
Features
Manage Home Assistant groups from chat
Selora AI can now create and maintain HA group helpers on your behalf. Ask it to group your living-room lights and it creates a real helper — the same kind you'd make in Settings → Devices & Services → Helpers — that you can edit yourself afterwards without touching any automation.
The point is that automations stop repeating long entity lists. They target one stable entity_id, and you change what's in the group whenever you like.
- Four new chat tools: list, create, update, delete.
- Groups are per-domain by construction; numeric sensor groups combine with a statistic (mean by default).
- Deleting a group goes through a confirmation card that names everything referencing it — automations, scripts, scenes, and parent groups, including legacy YAML groups — so you see the blast radius before confirming.
- Groups you defined in YAML are surfaced read-only, with an explanation, rather than silently duplicated.
Automation and scene creation is now animated
New automations and scenes reveal themselves in the panel as they're created, instead of appearing abruptly. The unused draft subsystem was removed along the way — the Automations tab shows live automations only, and pending proposals stay in chat where they belong.
Bug Fixes
Panel shell is clamped to the viewport
See the note above — this is the important one on HA 2026.8.0+.
Chat no longer stalls or leaks markup with cloud gateway models
Affects DeepSeek and other models served through a gateway. Three problems, one cause: these models sometimes write the tool call they want to make as ordinary text instead of issuing it properly, and Selora AI could not see it.
- The chat stopped mid-investigation. Because no tool call was visible, the turn was treated as finished — so a multi-step investigation just stopped, and you had to type "continue" to restart it. Selora AI now recognises what happened and asks the model to re-issue the call, so the turn carries on by itself.
- Raw markup appeared in the reply. Text like
<｜DSML｜tool_calls>reached the chat bubble. The filter that removes it only understood the plain ASCII form; these models use a lookalike character that renders identically. Both forms are now filtered. - The assistant repeated itself. While streaming, it lost track of what it had already said between steps, so it re-introduced the same sentence each round. Its own wording is now carried forward.
A restart-required banner instead of a cryptic error
When the integration is updated without restarting Home Assistant, the panel and the backend can fall out of step. That used to surface as extra keys not allowed @ data['…']. The panel now detects the mismatch and tells you to restart or reload, and the log gets a warning. Translated into all 13 supported locales.
Compatibility
No change to the minimum Home Assistant version (2025.1.0). The 2026.8.0 issue is a rendering regression this release fixes, not a new requirement.
Full changelog: v0.13.0...v0.14.0