fall-kit template forge

Consumes the foldkit-sdk and foldkit-mcp exemplars, then fills their shape for any estate tool via the fall-kit cascade — T0 off T2 WebLLM 1B–70B T3 BYOK Anthropic/OpenAI/Gemini. Alternate substrates: fallcompass fall-mcp-bridge fallcore.

Configure cascade

Open fall-kit settings to pick a tier. T0 works with no AI at all (mechanical template rename). T2 runs WebLLM in your browser — pick 1B (700MB) up to 70B (40GB). T3 uses your Anthropic / OpenAI / Gemini key, sent direct to the provider, never proxied.

tier: T0 · off (default) · click open cascade settings to change

Load exemplars

Fetches foldkit-sdk + foldkit-mcp raw files. These become the shape Llama fills.

idle

Choose input

Pick a source: a repo name to fetch from sjgant80-hub, direct source paste, or a queue of repo names for batch mode.

mode

repo name (without owner)

Configure

github token (optional · for private repos)

T2 WebLLM model (used only if tier=T2)

skip repos that already have -sdk / -mcp companions

Generate SDK

Streams Llama with structured prompt. Files: src/index.js src/index.d.ts package.json README.md examples/quick-start.mjs

idle

Generate MCP

Files: src/server.js src/tools.js src/resources.js package.json manifest.json README.md

idle

Download bundle

Creates two STORE-method zip files. Everything Llama produced, packaged for one-click download.

idle

Push instructions

Shell commands to unzip, create both GitHub repos, enable Pages, and tag v1.0.0. Review generated output before pushing.