Amira is your personal assistant that helps you make the most of artificial intelligence.
Integrated directly into HA, it allows you to create YAML cards, HTML dashboards, create automations, fix them, and integrate them with a single command.
You can work directly on the cards you need to modify or fix; it will automatically read the code, as well as automations, scripts, and helpers.
A floating bubble is always ready to respond to your requests, wherever you are.
It supports many LLMs with tokens and 4 LLMs with web subscriptions (OpenAI Codex, Claude, Perplexity, Gemini Web).
It also allows you to communicate with our HA via Telegram and WhatsApp.
This is an addon under development; I’m sure that with your suggestions and help in identifying bugs, it will become increasingly useful and stable.
Ai Assistant Update – Major Enhancements Beyond v3
Now is AMIRA
A new version of AMIRA is here — expanding multi-channel communication, adding web-based AI providers, and introducing local model support.
This release significantly extends what was possible in v3.
What’s New
WhatsApp & Telegram Integration
Amira now supports messaging platforms:
WhatsApp
Send and receive AI responses via WhatsApp
Webhook/API-based integration
Compatible with external providers (e.g., Twilio-style setups)
Telegram
Full bot support
Real-time command handling
Two-way communication with Home Assistant
This enables true conversational AI control from your phone.
MCP Model Context Protocol Server
Ollama Support (Local AI)
You can now run Amira with local models via Ollama.
Run fully offline
No cloud dependency
Choose models dynamically
Ideal for privacy-focused or self-hosted setups
This is a major step toward full local AI control inside Home Assistant.
Codex Web Integration
Support for Codex via Web API is now available:
Generate code on demand
Use web-based authentication
Execute structured developer prompts
Perfect for automation scripts, YAML generation, and development workflows.
Claude Web Mode
Direct connection to Claude via Web access:
Token-based authentication
No SDK dependency
Lightweight and flexible configuration
Faster setup compared to traditional API wrappers
This allows broader compatibility and simplified deployment.
Architectural Improvements
Better logging
Automatic provider fallback
Cleaner configuration structure
Improved compatibility with newer Home Assistant versions
More modular design
Feature Comparison
Feature
v3
New Version
Terminal Chat
WhatsApp
Telegram
MCP Routing
Ollama Local Models
Codex Web
Claude Web
Why This Matters
This update transforms Amira from a single-provider assistant into a:
Since the v3 era, Amira has evolved into a much more robust and production-ready Home Assistant AI companion. Here are the most interesting highlights from the major releases that followed:
What changed since v3
1) Full provider architecture rewrite
We rebuilt the provider layer to support a broad multi-provider ecosystem, including API-key providers and web-session providers, with better normalization, error handling, and fallback behavior.
2) OpenClaw-style core improvements (v4.5+)
A big milestone was the OpenClaw-inspired architecture:
Dynamic model catalog as a single source of truth
Real-time token and cost tracking
Multi-agent system with per-agent model/tool profiles
Stronger safety guards for write operations
3) Smarter model and fallback system
Fallback moved from a simple provider sequence to a much more controllable system:
Better fallback reliability and malformed tool-call recovery
Unified tool-calling normalization across providers
LLM priority configuration with clearer control over behavior
4) Dashboard generation quality jump
HTML dashboard generation received major upgrades:
Better design guidance and layout quality
Stronger auth/data-loading reliability
Improved handling for weaker/no-tool models
Better save/finalization logic for long or chunked outputs
Cleaner history behavior after dashboard creation
5) MCP improvements (runtime + UX)
MCP became far more usable:
Dynamic MCP tools correctly injected and executable in LLM-first flows
Better start/stop/status controls
Runtime persistence and improved server visibility in UI
6) Messaging and security hardening
Messaging support matured significantly:
Telegram whitelist for authorized user IDs
Better bot reliability and safer behavior
Continued WhatsApp/Twilio integration improvements
This project appears to be AI written, including all responding posts from @Bobsilvio. If you want this post restored, you need to have a disclaimer in this post as well as the github page that your entire library is vibe coded. There is evidence all throughout the code that this was not written by someone who uses AI as a tool.
I don’t see the problem. I’m a programmer, I’ve released at least 5 or 6 addons written by me, and this one is made with the help of AI, there’s no need to disguise or hide anything.
Where’s the difference for the end user?
Just a benefit.
You should also get up to speed, no vibe coding.
The modification is made both here and on GitHub, no problem, no desire to hide anything.
4.6.21 — xAI API, fallback/agent reliability, Ollama Cloud key support
New provider: xAI (Grok)
Added native xai provider with official endpoint https://api.x.ai/v1
New add-on config key: xai_api_key (runtime export XAI_API_KEY)
xAI is now available in the provider selector UI, model catalog, and persisted runtime selection
Fallback and orchestration
xAI integrated into the automatic fallback chain (priority, key detection, provider model override)
Full compatibility with legacy/enhanced ProviderManager and startup model cache refresh
Updated Grok models + readable names
Updated xAI model list with recent Grok models (e.g. grok-4-1-fast-*, grok-4.20-*, grok-code-fast-1, grok-3*)
Improved UI display humanization for xAI models not yet statically mapped
xAI default set to grok-4-1-fast-non-reasoning
Fallback model selector (UI)
Fixed fallback priority dropdown data parsing in Settings so provider model lists are correctly shown (instead of always Default provider model)
Added compatibility with both legacy and technical model payload shapes returned by /api/get_models
Custom agents runtime fix
Fixed active-agent runtime synchronization so selected provider/model is correctly applied after agent switch/reload
Improved persistence/re-init path to avoid mismatch between agent configuration and active chat provider/model
Gemini API-key schema fix
Fixed Google Gemini request schema issue that caused HTTP 400 with:
GenerateContentRequest...properties[disabled_by].enum[1]: cannot be empty
Grok Web cleanup (experimental)
Removed grok_web from public catalog/update flow
Added migration safeguard: if stale runtime selection still points to grok_web, it is automatically mapped to xai
Improved runtime warning/error clarity for removed provider state
Ollama Cloud API key support
Added new add-on config key ollama_api_key (OLLAMA_API_KEY export in run script)
Ollama provider/model fetcher now supports Authorization: Bearer <key> for cloud endpoints
Startup model refresh now passes Ollama API key for dynamic catalog refresh
Auto cloud host resolution: if ollama_api_key is set and ollama_base_url is left at local default, Amira automatically uses https://ollama.com (endpoints under /api/...)
Documentation
README and DOCS updated with xai_api_key and ollama_api_key usage notes
Thank you for making this extension. Unfortunately, I found it quite buggy. (e.g. some of my replies were in Spanish but my language was set to English, very “vibe-coded” feel to the entire UI). I did not feel comfortable with this, given the deep integration this extension requires.
Unfortunately, even after I uninstalled the extension, I still have leftover UI elements all over the place. One example is a chat bubble icon in my integrations menu.
To any future users looking to uninstall, I think I have fixed it by removing the /www/ha-claude-chat-bubble* files that were left behind. Please let me know if I have missed any other leftover references to this extension after install as I am looking to remove this extension completely.
Recent Updates
What’s New
WhatsApp & Telegram Integration
WhatsApp
Telegram
MCP Model Context Protocol Server
Ollama Support (Local AI)
Codex Web Integration
Claude Web Mode
Architectural Improvements
Feature Comparison
Why This Matters