Local LLM Tools¶
Optional role (installLlmTools). Install via chezmoi init --force and enable the prompt.
Related Notes¶
- OpenCode: local-only OpenCode config notes, known title-generation regression, and current workaround.
Tools¶
| Tool | Binary | Purpose | Install path |
|---|---|---|---|
| Ollama | ollama |
Local LLM runtime and API server | macOS: Homebrew formula or ollama-app via Brewfile; Linux: official installer |
| LiteLLM | litellm |
Local proxy / unified API for LLM backends | uv tool install litellm[proxy] --with httpx[socks] |
| llmfit | llmfit |
Recommend models that fit your hardware | macOS: Homebrew formula; Linux: official local installer |
| models | models |
Browse models, benchmarks, coding agents, provider status | macOS/Linuxbrew: Homebrew formula; Linux fallback: cargo install modelsdev |
Platform Notes¶
- macOS,
installBrewApps=false: ansible installsollama,llmfit, andmodelsvia Homebrew formulas. - macOS,
installBrewApps=true: ansible still installsllmfitandmodels, butollamais deferred to~/.config/homebrew/Brewfile.darwinasollama-app. - Linux with sudo: ansible installs
litellm,llmfit,models, andollama. - Linux with
noRoot=true: ansible installslitellm,llmfit, andmodels, and explicitly skipsollama.
macOS GUI vs CLI¶
ollama-appis preferred when you already opted into GUI apps withinstallBrewApps=true.brew install ollamais used when you want the CLI/runtime without the GUI app.- The macOS app also exposes the
ollamaCLI from inside the app bundle.
Common Usage¶
# Start the Ollama API server
ollama serve
# Run the LiteLLM proxy
litellm --help
# Check which models fit your hardware
llmfit
# Browse models and benchmarks
models
noRoot Reminder¶
noRoot=true skips ollama because the official Linux install path sets up system locations and a systemd service. The other tools remain user-level installs.