ai.nvim
A provider-agnostic ask/stream layer for talking to an AI from inside Neovim -- context assembly (buffer/selection/diagnostics), a streaming answer panel, and one :Ai command, instead of every plugin hand-rolling its own curl calls.
- status
- beta — Beta stage — active development. This repository is past its first shape and in active use, but the surface is not frozen: breaking changes are still possible. Pin a commit or tag if you depend on it.
- category
- AI
- neovim
- 0.11+
- repo
- github.com/WKDSteve/ai.nvim
- commits
- 184
- help
- :help ai
install
{
"WKDSteve/ai.nvim",
opts = {},
} also as
- native app: ai-desktop — Local agent-framework server (C++17, cpp-httplib).
docs/
- BINDINGS.md
- README.md
- architecture.md
- attachments.md
- bulk.md
- commands.md
- configuration.md
- health.md
- installation.md
- quickstart.md
- requirements.md
- scope.md
git log
- fix sheets: key sheet expects provider_resolution after reset, run lines name --task
- fix sheets: room for thinking in the model sheet, the key file removed on any exit
- test sheets: live sheets for the Claude API features, a manual one for the key setup
- test sheets: :Ai trust in the real viewer, blink.cmp, live Ollama and ai-desktop
- fix ollama: effort low switches a thinking model's thinking off
- test sheets: live sheet for num_ctx against a real Ollama
- fix ollama: send max_tokens as num_predict, warn about a long prompt once
- fix repo_prompt: write the trust store atomically, keep an unreadable one
- fix repo_prompt: a prompt file that links out of the project root is refused
- fix repo_prompt: :Ai trust allow in the review popup trusts the file shown
- feat the repository's own system-prompt file, trusted per content
- test sheets: room for thinking in the live provider sheets