◊·κ=1 · prime 1301 · part of the AI Native Solutions estate

One cascade. Every seed. Same level.

Drop-in JS module that gives any estate seed the AI cascade and P2P mesh users now expect — T0 off (default) · T2 WebLLM in-browser (1B → 70B) · T3 BYOK frontier (Anthropic / OpenAI / Google) · plus WebRTC mesh on the canonical fall-signal channel. Inlines for sovereignty. ~16 KB.

The 3 tiers

Tier	What you get	Where the work happens
T0 · off	The seed runs fully without AI. Default for every install.	Nothing downloads · zero API calls
T2 · WebLLM	In-browser LLM via WebLLM. Pick a model: 1B (700MB · any laptop) · 3B (2GB · balanced default) · 7B (5GB · capable) · 8B (5GB · common) · 70B (40GB · frontier, needs serious GPU + 64GB RAM)	100% in your browser · runs offline forever after first load · zero data leaves the device
T3 · BYOK	Bring your own key for frontier models. Anthropic (Sonnet 4.5 · Opus 4.7 · Haiku 4.5) · OpenAI (GPT-4o · 4o-mini · o1-mini) · Google (Gemini 1.5 Pro · 1.5 Flash · 2.0 Flash)	Direct fetch to provider · key stays in your browser · never proxied

Try it live

The chip is mounted at the top right of this page →

Click it to open the cascade settings. Then ask the kit to generate something with T0 (graceful null) · T2 (load a model first) · T3 (paste a key).

→ output will appear here

Embed in your seed

Two ways. Inline for sovereignty (recommended) — copy the contents of fall-kit.js inside a <script> tag in your HTML. Or reference via CDN if you trust this hosted copy:

<script src="https://sjgant80-hub.github.io/fall-kit/fall-kit.js"></script>
<script>
  FallKit.init({
    seedName: 'my-seed',
    chipAnchor: 'header .actions'  // CSS selector for where the AI chip goes
  });

  // Use anywhere you want AI assist · degrades gracefully at T0
  const draft = await FallKit.aiComplete(
    'You are a tight technical writer.',
    'Summarise this in 30 words: ' + sourceText,
    200
  );
  if (draft) showDraft(draft);
  else /* T0: fall back to your non-AI logic */;
</script>

Public API

Method	Returns	What
`FallKit.init({seedName, chipAnchor})`	`{version}`	Installs CSS · mounts AI chip · returns kit version
`FallKit.aiComplete(sys, msg, maxTokens)`	`string\|null`	Universal entry. T0 → null. T2/T3 → completion. Caller MUST handle null gracefully.
`FallKit.loadWebLLM(modelKey?)`	`Promise`	Lazy-loads WebLLM with chosen model (defaults to user's last pick or '3b')
`FallKit.openSettings()`	void	Opens the cascade settings modal
`FallKit.meshStart({seedName, onMessage})`	void	Activates BroadcastChannel mesh on `fall-signal`
`FallKit.meshPost(kind, payload)`	`bool`	Broadcasts to other seeds on the channel
`FallKit.helpSection()`	`string` (HTML)	Returns the standard help-section HTML to inject into your seed's Help tab
`FallKit.MODELS`	object	Registry of available WebLLM models with size + label
`FallKit.PROVIDERS`	object	Registry of T3 BYOK providers + their models

Sovereignty doctrine

T0 must always work — the kit never gates a seed's core feature behind AI. aiComplete returns null; callers fall back to non-AI logic.
API keys never proxied — direct fetch from the user's browser to the provider. Key stored only in localStorage.fall-kit.config. Never logged, never sent to us.
Lazy-loaded WebLLM — model weights download only on explicit user opt-in. Kit itself is ~16 KB.
BroadcastChannel mesh on fall-signal — the canonical channel name across all estate seeds (consistent with fallnet).
Estate-first canonical — WebLLM pattern ported from botler. WebRTC channel from fallnet. Konomi 7-prime stack consistent. ◊·κ=1.

What ports cleanly

Any seed already using single-HTML pattern with IndexedDB and an existing tab/nav
Drop into the head <script>, call FallKit.init() on boot, wire aiComplete into your assist features
Help tab? Add helpEl.innerHTML += FallKit.helpSection() — consistent across the estate

◊ Fall Kit v1.2.0 · MIT · github.com/sjgant80-hub/fall-kit · part of the AI Native Solutions estate · ◊·κ=1