jwulf/new-mac-setup

0. Install Karabiner and set up Halmak layout
1. Install nvm
2. Install Node 24: nvm install 24
3. Install TypeScript: npm i -g typescript
4. Install brew: /bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"
5. Install zed
6. Install Haskell via GHCUp: curl --proto '=https' --tlsv1.2 -sSf https://get-ghcup.haskell.org | sh
7. Install GitHub CLI: brew install gh
8. Install GitHub Copilot: npm i -g @github/copilot
9. Install Cocktail: npm i -g @camunda8/cli@alpha
10. Install Pi: npm install -g --ignore-scripts @earendil-works/pi-coding-agent
11. Install Pi LSP: pi install npm:pi-lsp-extension
12. Install Pi GitHub: pi install npm:pi-gh-cli
13. Install Pi llama.cpp: pi install npm:pi-llama-cpp
14. Install Pi Context-Mode: npm install -g context-mode && pi install npm:context-mode
15. Install Pi Memory: pi install npm:@pi-unipi/memory
16. Install Zed Biome: "Open zed: extensions and search for Biome"
17. Install Zed Haskell: "Open zed: extensions and search for Haskell"
18. Install llama.cpp: brew install llama.cpp
19. Get the Gemma 4 model (from here): llama-server -hf unsloth/gemma-4-26B-A4B-it-GGUF:UD-Q4_K_M --port 1234 -ngl 99 -c 32768 -np 1 --jinja -ctk q8_0 -ctv q8_0
20. Configure pi for llama.cpp — edit ~/.pi/agent/settings.json add: "llamaServerUrl": "http://127.0.0.1:1234"
21. Generate SSH Key for GitHub
22. Install OpenJDK: brew install openjdk
23. Install GhostTTY
24. Install Orbstack: brew install orbstack
25. Install uv (Python): curl -LsSf https://astral.sh/uv/install.sh | sh
26. Install asdf: brew install asdf
27. Install asdf .NET plugin: asdf plugin add dotnet
28. Install .NET 8: asdf install dotnet 8.0.421
29. Install .NET 9: asdf install dotnet 9.0.314 30: Install .NET 10: asdf install dotnet latest
30. Install Clojure: brew install clojure/tools/clojure
31. Install Zed Clojure: "Open zed: extensions and search for Clojure"
32. Install Little Coder: npm i -g little-coder
33. Set nano as git editor: git config --global core.editor "nano"
34. Set up git identity: git config --global --edit
35. Install bun: npm i -g bun
36. Install qmd for Pi memory search: npm i -g @tobi/qmd
37. Install MacMLX, and download Qwen model
38. Install omlx: brew install omlx --with-grammar
39. Model for omlx config in ~/.config/little-coder/models.json:

{
  "providers": {
    "omlx": {
      "api": "openai-completions",
      "baseUrl": "http://127.0.0.1:8000/v1",
      "apiKey": "IGNORED",
      "models": [
        {
          "id": "Qwen3-32B-4bit",
          "name": "Qwen3.6-35B-A3B (local omlx, 150K)",
          "reasoning": true,
          "input": ["text"],
          "contextWindow": 150000,
          "maxTokens": 4096,
          "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }
        }
      ]
    }
  }
}

Serve several local GGUF models behind one OpenAI-compatible endpoint using llama.cpp's built-in router mode. This exposes llama.cpp's native model management (/models, /models/load, /models/sse) that the pi-llama-cpp extension uses to browse/load/switch models — plain llama-swap does not implement those endpoints (pi gets HTTP 404), so use the router for pi.

./start-llm-router.sh          # binds :: (IPv4+IPv6) on :8888

• Models are defined in llama-router.ini (currently qwen3.8 and minicpm5-2b); the [*] section holds shared defaults. Local models use model = /path/to.gguf, remote ones use hf = user/repo:quant.
• The router also auto-lists any other GGUFs in your Hugging Face cache.
• Env overrides: LLAMA_ROUTER_PORT, LLAMA_ROUTER_HOST, LLAMA_ROUTER_MODELS_MAX (default 2 = both resident; 1 = hot-swap).
• Clients pick the model via the OpenAI model field, e.g. qwen3.8 or minicpm5-2b. pi: set llamaServerUrl to http://<host>:8888; little-coder: set baseUrl to http://<host>:8888/v1.
• LAN clients that resolve <host>.local to an IPv6 link-local address are covered by the dual-stack (::) bind; if a client still fails, use the IPv4 literal (e.g. http://192.168.0.141:8888).

start-llm-qwen-3.8.sh remains a standalone single-model launcher with KV-cache save/restore (takes an optional port arg); the router does not use it.