CAAL functionality with Home Assitant Voice Assistant

[finnaGIT]

First off, huge shout out to @cmac86 for this project.

I'm wondering how i can generalize some of the exciting features that CAAL offers, but using home assistant's voice assistant? recognizing this is vague, i'll describe what i mean:

i currently have home assistant with the voice assistant set up (using faster-whisper kokoro-FastAPI both with GPU inference), ollama for the conversation agent, and n8n running.

i'd love to be able to tell the voice assistant "create an n8n workflow that gets NFL scores." or, use the home assistant voice assistant to make other tool calls in n8n.

[cmac86]

@finnaGIT, Thanks for the kind words!

Are you using Ollama as a conversation agent in HA, or just the basic Assist intents?

If Ollama is in the loop, theoretically you could connect HA to n8n via MCP and expose workflows as tools. But there's no tool discovery - you'd have to manually wire each workflow. And HA's conversation architecture doesn't have persistent context or follow-up handling. Many things different about the architecture.

That's honestly why I built CAAL separately. HA is great for device control - CAAL just treats it as one tool among many.

My next video is literally about this - why I moved away from HA Assist and built CAAL.

[finnaGIT]

thanks for the response! look forward to the next video.

i'm using ollama as a conversation agent.

regarding tool discovery - fair point. CAAL is unique there. i'll have to chew on this myself. i thought that n8n agents had the capability to do discovery (i.e., looking across its own workflows to make the right tool call) but haven't given it much thought. EDIT: or, have a primary n8n agent with all tools wired to it, one of which is a tool to modify the primary agent to add new/revise existing tools.

good point that there's not persistent context. that said, a contextually-relevant tool call from HA to n8n (something like a 'remember this' -> 'remind me about...' type of flow), i think i could leverage n8n's agent and attach a vectordb and let ollama handle embeddings there.