🤖 Local AI in Neovim

Configured via CodeCompanion.nvim talking to a local OpenAI-compatible endpoint. On-demand — no daemon running unless you start it.

flowchart LR
    A[nvim<br/>&lt;leader&gt;aa] -->|chat| B[CodeCompanion]
    B -->|/v1/chat/completions| C[mlx_lm.server :8080]
    B -->|/v1/chat/completions| D[LM Studio :1234]

Two ways to serve a model

🅰️ MLX server (CLI)🅱️ LM Studio (GUI)

# 1. Drop in ~/.zshrc.local
export MLX_MODEL_PATH="/path/to/gemma-3-moe"

# 2. Start the server
mlx-start                # :8080

# 3. Point CodeCompanion at it
ai-use-mlx

# 1. Open LM Studio app
# 2. Load your Gemma 3 model
# 3. Start the local server (defaults to :1234)

# 4. Flip CodeCompanion's endpoint
ai-use-lmstudio

Helpers

Command	Effect
mlx-start [model]	Launch mlx_lm.server on :8080
mlx-stop	Kill any running mlx_lm.server
mlx-status	Is it up?
ai-use-mlx	Set $AI_LLM_URL → :8080 for current shell
ai-use-lmstudio	Set $AI_LLM_URL → :1234 for current shell
ai-status	Print the current endpoint + ping it