A bit more documenation would be useful to get started! (a few questions after trying it)

[jemisa]

Looks interesting!

Can you explain how the -dir folder work ? when I try to run it with -dir myproject, it search files like ./main.go and not ./myproject/main.go ?
Can you also explain how -output option is supposed to work ? right now, I found out it forces the writing of the file (summarizing the project) and does not display the TUI.

You might want to document q/exit in your README

Can you also add a sample usage how to use ASK then AGENT mode ? On my terminal, the ASK/AGENT bolding is not really working... Maybe be add a AGENT ?

Is there a way to alternate between ASK and AGENT mode while running ? (like Control-Y activates/deactivates YOLO mode)

Anyway, I manage to get it to work with llama.cpp,
here is my config file:

{"llm":{"api_key":"none",
        "model":"default",
        "endpoint":"http://192.168.100.1:8081/v1/chat/completions"
       }
}

I used the option
-config ~/.local/vogte/config.json
when starting vogte

Every message is considered a new chat and not related to the previous.

Do you think supporting something like AGENTS.md that would be included in each message would help,
so I don't have to repeat a few information each time?

Or at least make some history working, so I can tweak the prompt

[piqoni]

Hi thanks for showing interest.

1. Things will change (but added some more docs, thanks) but -dir is simply redundant (as long as you just execute vogte binary in the relevant directory you are interested, you should never need to use -dir).

I used it for development when I wanted the project structure of other projects(dirs) in my current dir, like so:

2. -output is on "cli" mode, no TUI. Its job is simply to generate the project structure in case you need it to put the context yourself in some other model or ui. See above picture how to generate it.

3. May I know your terminal. I've tested with Iterm2 and Alacrity and clicking on agent/ask works. To start in agent mode simply run: "vogte -agent" and agent mode should be bold.

Thanks for testing it out and for the suggestions, they are reasonable. One tip that might alleviate in terms of history is that vogte should create a hidden file /.vogte/chatbot.log with the history and LLMs response.

[jemisa]

3. My terminal is WezTerm (over ssh if it matters):

Since you mentionned I can click to change mode, after I click once on either ASK or AGENT, the 'bold' is visible.
But before that, none of the two are bolded.

I tried both way (with or without -agent )
~/go/bin/vogte -config ~/.local/vogte/config.json -agent
or
~/go/bin/vogte -config ~/.local/vogte/config.json

In both cases, I need to click once on ASK or AGENT for it to become 'bold'.
But vogtet does know which mode it is on, because if I am in -agent, clicking on Agent does not change anything, but clicking on ASK displays "Mode set to ASK"
if I click outside of ASK or AGENT (e.g. on the model name), it unbolds either ASK or AGENT.