Pipeline Instrument

Single-file browser instrument for AI investment pipeline work. It builds two offline HTML artifacts from one vanilla source tree:

• dist/instrument-practice.html: practice tier with local OpenAI-compatible LLM features.
• dist/instrument-client.html: client tier with LLM code stripped at build time and no network calls.

Build

npm run build
npm run check

The check verifies the file-size ceiling and confirms the client artifact does not contain practice-tier LLM tokens.

Use

Open either HTML file directly from disk in Microsoft Edge. No local server is required.

Portfolio JSON is canonical. The instrument uses the File System Access API when available for save-in-place, with download and file-input fallbacks. A browser recovery snapshot is stored in localStorage every 60 seconds and on tab hide; use Clear this device inside the app to remove those snapshots.

Practice-tier LLM endpoint

The practice tier calls an OpenAI-compatible chat completions endpoint, defaulting to http://localhost:11434/v1.

For Ollama from a file:// origin, start Ollama with:

$env:OLLAMA_ORIGINS="*"
ollama serve

The client build contains none of the LLM UI, endpoint configuration, prompts, or fetch calls.