Claude Chat for Home Assistant — chat panel + dashboard editing with diff approval

Share your Projects! Custom Integrations
, ,
1

Claude Chat for Home Assistant — chat panel + dashboard editing with diff approval

Hey everyone :waving_hand:

I’ve been hacking on a custom integration that adds a Claude Chat entry to the HA sidebar. The idea: a real chat panel (multiple sessions, like ChatGPT) where I can say things like “add a card to my test dashboard with the living room temperature as a gauge” and Claude generates the change, shows me a unified diff, and only applies it after I click Apply.

It’s different from HA’s built-in anthropic integration, which is a voice/Assist agent. This one is a dashboard authoring co-pilot — its tools are wired into Lovelace storage and HA service calls, with a hard “diff + approve” gate so nothing changes without a click.

Repo: GitHub - h0jeZvgoxFepBQ2C/claude_chat_for_homeassistant: A sidebar chat panel for Home Assistant powered by Anthropic's Claude — inspect entities, edit dashboards, and create automations in plain English with diff-and-approve gating. · GitHub

What it does

  • Sidebar panel with persistent, auto-titled chat sessions

  • Streaming responses (markdown, code blocks, tables render live)

  • Model picker per session (Haiku / Sonnet / Opus), choice is remembered

  • Tools Claude can call:

  • list_entities, get_entity, list_areas

  • list_dashboards, get_dashboard, list_lovelace_resources

  • list_automations, get_automation, list_services

  • propose_dashboard_update, propose_service_call, propose_automation_create / update / delete — all stage changes for your approval

  • Pending changes show a unified diff for dashboards, a payload preview for service calls, or the full YAML for automations. Apply or Reject from the chat UI. On approve, automations are appended to automations.yaml and automation.reload is called.

Screenshots

(see the GitHub README for full-size — TL;DR: empty state with example prompts, sensor query showing a markdown table from list_entities, dashboard diff + Apply/Reject)

Heads-up — first custom component

This is my first HA custom component. I’d really appreciate it if folks who know the HA internals could glance at the code and tell me anything I’m doing wrong, anti-patterns, places where I’m using deprecated APIs, etc. Areas I’m least sure about:

  • The way I read/write Lovelace dashboards via hass.data["lovelace"].dashboards — is there a more correct/forward-compatible API?

  • I had to wrap AsyncAnthropic(...) in async_add_executor_job because the SDK loads SSL certs synchronously and HA’s strict mode flags it. Is there a cleaner pattern?

  • Custom panel registration via frontend_url_path + module_url static path — feels right, but I’m not sure about long-term stability.

  • For automation create/update/delete I’m appending to automations.yaml (atomic write via temp file + os.replace) and calling automation.reload. This requires the default automation: !include automations.yaml line. Storage-mode-only automations (no YAML backing) aren’t editable yet — should I be using homeassistant.components.config.automation websocket commands instead?

Privacy

The integration only sends what tools fetch (entity lists, dashboard configs, etc.) to Anthropic — it does not stream your full HA state. Sessions are stored locally in .storage/claude_chat.sessions. You bring your own Anthropic API key.

Status

Experimental — works, but new project and not fully tested. Don’t use on a critical instance yet. Issues / PRs / forum replies very welcome.

Cheers!

1 Like