README.md

metadata

title: Local Frontier
sdk: docker
license: mit
short_description: Interactive local AI hardware and model-bound database.

Local Frontier

Interactive database for local AI hardware notes plus model metadata and theoretical memory-side decode bounds.

Model Data

• assets/local-frontier-model-data.js is the model catalog loaded by the static app.
• schemas/local_frontier_model.schema.json documents the durable model shape.
• scripts/validate-model-data.mjs validates required model, architecture, adapter, and source fields.

Validate after editing model metadata:

node scripts/validate-model-data.mjs

Validate the bounds calculator reason codes:

node scripts/test-bounds-engine.mjs

TypeScript UI Migration

The current Space still serves app.html through server.py; do not add an index.html. New React/shadcn migration work lives under src/ and is checked with Slophammer.

Run the full local gate:

npm run check
npm run coverage
npm run dry
npm run slophammer
npm run mutate

Compare URLs

The model/hardware comparison page supports stable deep links with one or more models and one or more hardware rows:

/compare?models=<huggingface-namespace%2Fmodel-id>[,<huggingface-namespace%2Fmodel-id>...]&hardware=<hardware-id>[,<hardware-id>...]

Examples:

/models/Qwen/Qwen3.6-35B-A3B
/compare?models=Qwen%2FQwen3.6-35B-A3B&hardware=nvidia-dgx-spark
/compare?models=Qwen%2FQwen3.6-35B-A3B,google%2Fgemma-4-26B-A4B&hardware=nvidia-dgx-spark
/compare?models=Qwen%2FQwen3.6-35B-A3B&hardware=nvidia-dgx-spark,apple-macbook-pro-16-m5-max-128gb
/compare?models=Qwen%2FQwen3.6-35B-A3B,google%2Fgemma-4-26B-A4B&hardware=nvidia-dgx-spark,apple-macbook-pro-16-m5-max-128gb

Repeated query params are also accepted. Model detail pages use the Hugging Face repo id directly under /models/. Older URLs such as /bounds/<model-id>/on/<hardware-id>, /bounds/<model-id>, and /bounds?model=...&hardware=..., plus prior slug model IDs, remain accepted and are normalized to canonical Hugging Face IDs in the browser.