CodeCompanion & DAP #1784
Replies: 1 comment
-
I've just added a new tool that allows the LLM to step into/out/back... I tried it with debugpy and gemini2.5-flash, and the LLM understood the tools surprisingly well, so well that I think Google might have included stuff like LSP and DAP in their training corpus so that they inherently have a pretty good idea about how to use these tools. @olimorris, as the tool ecosystem around codecompanion grows bigger, I seriously think that we need some sort of throttling for the adapters 🤣 I tried asking the LLM to debug through a Python process while keeping track of some variables, and quickly hit the rate limit. An async waiting mechanism that holds off a request until |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
-
Inspired by #1696, I thought it might be fun to expose some of the nvim-dap functionalities to codecompanion.
With the help of codecompanion itself, I've come up with codecompanion-dap.nvim, which (currently) exposes 5 of the readonly DAP requests to the LLM.
With codecompanion-dap.nvim, the LLM can now tap into a running DAP session and inspect the internal state of the process. This is still quite limited, and I plan to give it more capabilities such as breakpoints, step over/into/out, etc. and see what the LLMs are capable of.
This extension is still in a very early stage, and I've only tried it with debugpy and coredumpy. If you're a nvim-dap user, I'd love to hear your opinions on this!
Beta Was this translation helpful? Give feedback.
All reactions