§ live demo · cross-seed mesh in your browser

Two FallSeeds. Talking. Right now.

Both iframes below load from sjgant80-hub.github.io. They're same-origin to each other, which means they share a BroadcastChannel('fall-signal'). They discover each other in milliseconds, no server, no account, no handshake. The mesh is just the substrate.

Open both seeds Each iframe loads in ~3 seconds. They announce themselves on the mesh on boot.

Click "Mesh" or "Inspect" in either iframe — you'll see "Fellow seeds online" listing the OTHER iframe with its prime, version, level. Live, updated every 30 seconds.

In fallrecall, click "load sample" below to drop a tiny mock conversation. Then in fallrecall's Analyse tab, click "Analyse now" (needs a free LLM provider configured first, ~30 seconds).

two iframes · same origin · BroadcastChannel discovery

fallrecall · the exit ramp sjgant80-hub.github.io/fallrecall/

fallseed-creator-os · L2 personal-OS sjgant80-hub.github.io/fallseed-creator-os/

Note on iframes vs tabs: in iframes the seeds can mesh-talk to each other (same origin), but some browsers block localhost:11434 autodetect from iframes, and IndexedDB scopes are per-origin so they share the same DB. For the full demo (every feature, including private IDB per seed and local-ollama autodetect) click open both in separate tabs above — same mesh, cleaner environment.

Sample flow · pull flavour from fallrecall into creator-os

1. In the LEFT iframe (fallrecall) → click Import → click Generic — paste any conversation → paste a couple of paragraphs of something you've written → save. Or use the "load sample" button below to inject a tiny mock conversation.

2. In the LEFT iframe → click LLM Providers → enable at least one (Groq is fastest free tier · 30-second signup at console.groq.com/keys).

3. In the LEFT iframe → click Analyse → Analyse now → wait ~30 seconds → profile lands.

4. In the RIGHT iframe (creator-os) → click Inspect → scroll to "Flavour from fallrecall" → click ⌬ pull flavour. Profile lands instantly — no server, no API, no account. Pure mesh.

5. In the RIGHT iframe → use the Agents tab — every cascade call now auto-prepends your flavour as the system prompt prefix. The persona's voice = your voice.

~80 user messages · 4 topics · ready for analyse

What this proves

1. The mesh is real. Two seeds sharing an origin discover each other in <100ms via the browser's BroadcastChannel API. No protocol negotiation, no auth, no DNS.

2. The protocol is open. The request/response handlers (introspect · records-export · flavour-summary) are defined in the public spec. Any new seed satisfies them and instantly composes with the existing estate.

3. Sovereignty composes. fallrecall captures your AI flavour locally. Creator-OS pulls it via the mesh. Your data goes nowhere. The "platform" effect — your AI experience getting better because it knows you — happens entirely on your device.

4. The labs aren't required. Once your flavour is captured and stored locally, every cascade call in creator-os or agents prepends it as a system prompt. WebLLM (in-browser) or ollama (local) means the lab never sees your data, even during use.

Read the spec for the formal protocol →