Undergoing major experimental changes for now
AI plugin for Vim/NeoVim with focus on local evaluation, flexible context and aggressive cache warmup to hide latency.
vimqq_thor_dst.mp4
Features (including experimental)
- Support for both remote models through paid APIs (Claude, Deepseek) and local models via llama.cpp server;
- automated KV cache warmup for local model evaluation;
- dynamic warmup on typing - in case of long questions, it is a good idea to prefill cache for the question itself;
- human-readable hierarchical project indexing;
- llm agents in different roles: engineer, reviewer, etc.
- fully closing the loop and implementing complex features E2E