How does the cmp integration work?

[mrjones2014]

I see a cmp_codecompanion module in the source tree but I can't find any docs on it

[olimorris]

Yeah it's not too fleshed out yet. You'll see if you try to change a model in the settings it will offer completion based on the adapter's schema:

[mrjones2014]

Ah okay I thought it would like produce completions via AI like Copilot or something.

[olimorris]

Oh I've given that some thought...

With Ollama, it makes a lot of sense...with Anthropic or OpenAI, it's super expensive as I imagine it's just the constant sending of tokens to the LLM.

I am keen to work on a feature where you could send the whole code base over to an LLM and ask for a specific feature or bug fix. For an open source plugin that could be really cool

[mrjones2014]

I am keen to work on a feature where you could send the whole code base over to an LLM and ask for a specific feature or bug fix.

This sounds like what cody is. We have it at work, I believe it runs in our own self hosted infrastructure so that it can use our proprietary code as context.

[olimorris]

Yes I'd heard of Cody but never tried it out. Copilot Chat has a similar feature now I believe.

Would love to unpick how Cody goes about doing it.