Commit

The repo is live and public at https://github.com/slogsdon/llm-monitorslogsdon/llm-monitor. Your local main matches origin/main at 55206f0, and the tests pass (27 passed, 0 failed). - **Description:** "Terminal dashboard and JSON reports for a local LLM stack on macOS: LiteLLM, llama-swap, oMLX/mlx_lm, Ollama" - **Topics:** apple-silicon, cli, dashboard, litellm, llama-cpp, llama-swap, llm, local-llm, macos, mlx, monitoring, ollama, tui, typescript - **Commits:** - 941f063 fix: drop machine-specific LiteLLM config default: removes the ~/Code/otel-local-ai/... fallback from src/config.ts:24. - 55206f0 docs: humanize README, add license and package metadata: adds a LICENSE file (MIT, which package.json already declared but no file existed), plus keywords, repository, homepage, bugs and author in package.json. **Two things you need to act on:** - **Set LITELLM_CONFIG yourself.** It isn't set in ~/.zshrc or ~/.zshenv. Without it, the dashboard on this machine no longer finds your LiteLLM config, and Routing falls back to empty. Add this to your shell config: export LITELLM_CONFIG=~/Code/otel-local-ai/litellm/config.yaml. - **The old path is still public.** The first three commits contain otel-local-ai and went up with the push. Getting rid of it would mean rewriting history and force-pushing, so I haven't done that. Say if you want it. **README humanize pass** **Score:** 100. The first pass scored 92. **Deductions remaining:** none. The — in Known gaps is the dashboard's own placeholder character inside a code span, not punctuation. **Changed lines:** it's close to a full rewrite, so here are the main changes. The complete diff is in git show 55206f0 -- README.md. - Intro: "Polls every provider concurrently, degrades to OFFLINE instead of crashing…" → "Every provider is polled concurrently. One that's down shows as OFFLINE and the rest of the screen keeps working." - First-pass intro "shows what's loaded, what's generating and how fast, and tells you when something has gone down" → "shows which models are loaded and how fast they're generating, and flags any server that has gone down" (it lost points as a list of three). - Install: cd ~/Code/llm-monitor → git clone https://github.com/slogsdon/llm-monitor.git then cd llm-monitor - "Requires Node ≥ 23.6 (tests run TypeScript directly via native type stripping)." → "You need Node 23.6 or newer. The tests run TypeScript directly through Node's native type stripping." - Resource use: the bold-label bullets ("**Bounded V8 young generation.** …", "**Cached port discovery.** …") → plain sentences. Every em-dash and semicolon in the prose became a period or a comma. - "…the shim is executed directly, if you launch via node dist/cli.js…" → "This only applies when you run the llm-monitor shim directly. If you launch with node dist/cli.js…" - "Organised by attention, not by provider: system meters on top, …" → "The screen is organised by what needs your attention, not by provider. System meters sit on top." - Ollama row: "context length, **opt-in** via --provider ollama (benchmarking sidecar…)" → "context length. Opt-in with --provider ollama, since I run it as a benchmarking sidecar rather than part of the serving stack" - The exit codes paragraph moved up under Usage, and Resource use moved below System metrics. - Added a npm run typecheck line under Development and a License section. My first draft also had a "why I built it" paragraph about curling four endpoints on a Mac mini. I made that up, so I took it out before publishing.