Talk to an Anam avatar inside Home Assistant. Ask about your exposed devices, or opt in to Assist controls.
Anam handles the conversation and avatar. Home Assistant runs the supported tools through your signed-in browser.
Experimental · Unofficial · MIT licensed
“What’s the temperature in the living room?”
You speak or type into the Anam Assist panel.
The conversation requests an available state-reading tool.
Checks the session and reads Assist-exposed state.
“It’s 21 degrees in the living room.”
Illustrative data. This page doesn’t start a session, use your microphone, or connect to your home.
Manual installation is the verified route for this preview. Back up Home Assistant before adding a custom integration.
Download source ZIP
Download and extract the repository. Copy the entire
custom_components/anam_assist folder into your
Home Assistant configuration’s
custom_components directory. Create that
directory if needed.
config/
└── custom_components/
└── anam_assist/
├── manifest.json
├── __init__.py
└── …The built browser bundle is included. You don’t need Node.js on your Home Assistant host.
Restart Home Assistant. Go to Settings → Devices & services → Add integration and search for Anam Assist.
Enter your Anam API key in Home Assistant. Add a published persona ID, or leave it blank for the public Cara example. The saved persona is not modified: this integration supplies its own prompt and tools for each session.
Portrait is the default, at 768 × 1152. Use a portrait-capable Cara 4 avatar, or choose landscape in Reconfigure for an older avatar.
Review the entities you expose to Assist. Open Anam Assist in the sidebar, optionally select a screen location, and start a conversation. Ask about an exposed sensor. Turn off Use microphone to test by typing.
End the session, choose Lights only under Home controls, and start again. Ask for one exposed light by name. Check the physical result, then end the session when you’re done.
Repository metadata is included, but the full HACS install/update/removal lifecycle has not been validated. This project is not in the default HACS catalog. Use manual installation for the verified path.
If you want to test the custom-repository route, add
https://github.com/Smidge/anam-assist with type
Integration, following the
HACS custom repository guide. This is a custom integration, not a Supervisor app/add-on or
a Lovelace card.
The browser connects the two. Your existing Assist pipelines stay in place; this adds a separate conversation panel using Home Assistant’s Assist LLM API.
The browser uses the Anam JavaScript SDK for microphone audio, typed input and avatar playback. Anam handles speech recognition, the conversation model, voice and video.
Home Assistant creates a short-lived session token. The Anam API key stays in the server’s config entry and normal HA backups.
Anam requests a client tool. The browser forwards it over Home Assistant’s authenticated WebSocket connection. The server checks the session, tool allowlist and arguments, then invokes the Assist API.
The result goes back to Anam before it answers. There is no public webhook, separate long-lived HA token, or HA server media relay.
| Task | Support |
|---|---|
| Read exposed state and live context | Available in read-only sessions |
| Switch lights and set brightness | Opt in before each session; exposed lights only |
| Media, fans, locks, covers, climate, switches, scripts and scenes | Available through Assist controls when Home Assistant supplies the tools. Review entity exposure first; actions have no separate confirmation. Exposed scripts and scenes may affect other devices. Broader physical-device testing is still pending. |
| Authenticated kiosk browser | Can open the panel; unattended recovery is not implemented |
| Wake words, casting and RTSP | Not supported |
This is a cloud-connected conversation. Anam receives audio or typed messages, Assist prompt context and tool results.
Live context can include all exposed entity states, even if you ask about just one sensor. Review Assist exposure before connecting.
Read the security modelSession video recording and stored transcripts are disabled. That does not mean zero provider retention. Local transcripts remain in page memory.
Light changes can trigger your existing automations. Ending a session cannot undo a command already sent to a device.
Evidence for this version, without a production-readiness claim.
View CI results57 backend tests and 40 Chromium/WebKit browser tests passed, alongside lint, a reproducible browser build and a runtime dependency audit.
Live testing on Home Assistant 2026.10.3 verified a portrait stream, a temperature query, and an exposed light turning off and back on. The owner confirmed both physical transitions. The final state matched the starting brightness.
Full clean-install HACS lifecycle testing, broader HA compatibility, and microphone/audio checks on physical mobile browsers. The live light check used typed commands. There has been no independent security audit.
For bugs, include your HA version and a synthetic reproduction. Keep API keys, home-state payloads and real transcripts out of public issues.
Report a bug
Check that the folder is exactly
custom_components/anam_assist inside your HA
configuration directory, with
manifest.json directly inside it. Restart Home
Assistant and refresh the browser.
Open Home Assistant over HTTPS and allow microphone access when you want to speak. Try a typed session with Use microphone unchecked. If the browser blocks playback, use the panel’s playback control.
For an older avatar, try landscape through the integration’s Reconfigure action. Check the Anam account’s streaming access and usage allowance.
Check that the entity is exposed to Assist and use its exact name. Read-only sessions cannot change devices. Even with light control enabled, unsupported device types remain unavailable.
An action may have completed even if its result timed out. Check the device’s state before asking again. After a disconnect, start a new conversation; an old session cannot transfer control to a new connection.
End active sessions, back up your configuration, replace the whole component, restart Home Assistant and refresh the browser. Update frontend and backend together.
To remove it, delete the integration through Devices & services before removing its folder. Your existing Assist pipelines remain unchanged.
No. The integration’s source is MIT licensed, but Anam supplies the cloud conversation and media. You need an Anam account and streaming usage. See Anam for current account options.
Start read-only, keep exposure deliberate, and tell us what you find.
Read the setup guide