Feature Request: AI integration for auto-replies / AI mode

[shingareom]

Hi, I’d like to propose a feature idea for nchat:

It would be really useful if nchat had an AI integration that can:

• Suggest or generate replies automatically,
• Optionally be enabled for specific contacts,
• Be toggled on/off with a simple keybinding.

This would make nchat even more powerful by reducing repetitive typing and allowing smart auto-replies directly inside the terminal interface.

I’m not proposing a solution or implementation right now... just the idea for consideration.
Would love to hear thoughts from the maintainers and community.

[d99kris]

Hi @shingareom - I think a feature that suggests / generate replies automatically could be useful. I have thought about this before too, so I'll probably try to look into this.

As for automatically also sending out the auto-generated messages, I'm a little hesitant, as non-interactive control (incl. sending of messages) is generally out of scope for this project.

[d99kris]

Hi again - support for auto-composing message using LLM services has been added with #403 / 891b455.

It's disabled by default and can be enabled in ui.conf by setting auto_compose_enabled=1

Additionally the environment variable OPENAI_API_KEY shall be set up with a valid API key for OpenAI. To use other LLM services, see Auto-Compose documentation.

Use the function by pressing Alt/Opt-i. With a message selected, the auto-compose tool will get access to only that message and older (up to max as specified by auto_compose_history_count). With no message selected, the tool will have access to the latest messages.

Example screenshot:

[bonswouar]

Pretty cool to have this compose feature integrated in nchat!
I've noticed it seems to only support "base" llamacpp api for local llm, not full OpenAI compatible endpoint.
Would it be possible to add an URL param to OpenAPI type directly?
I personally use llama-swap to expose OpenAI compatible endpoint with many models, thus the model parameter would be needed.
EDIT: Never mind, the -m parameter does work as expected with custom endpoint, cool!

Another related feature request, maybe a way to customize some other parameters? i.e prompt / system prompt, max tokens, etc

[d99kris]

Another related feature request, maybe a way to customize some other parameters? i.e prompt / system prompt, max tokens, etc

That can be done. Added support for custom prompt and max tokens limit in 61a523d / v5.12.13 - refer to Auto-Compose documentation for an example at the bottom of the page.

[d99kris]

Since the basic functionality for this feature request has been implemented, I will proceed to close this discussion.