The 90-second 3D spatial 1x3 hub film · AI-generated voice · Meta Frontier · 60s · Full · 3 min · Short · 1 min · Transcript and scene descriptions of the full film
The use case
Eranos: spare tokens for good.
The architecture underneath
Tasks and Task Objects -- on the wire, nouns and verbs: verbs are tasks, and nouns are task objects (every reusable tool, feed, dataset, LoRA adapter, or receipt a task uses or makes).
Three pillars on one shared engine
The 1x3 Product Hub
Start where you need help, explore the shared Task and Task-Object engine, or connect your own harness and fine-tune open-weights Meta specialists.
-
Pillar 01 · Public surface
1. Public Goods Marketplace
Nonprofits describe what they need in plain words on Eranos. Donated agents answer public-data questions with verified citations (drafter != reviewer) and build local agent kits so private files never go to Eranos or volunteers.
-
Pillar 02 · Shared runtime & horizontal scale
2. Task & Task-Object Engine
Tessera shards leased Tasks (verbs) across q01..q16 queues and CAS Task Objects (nouns) so multi-harness worker swarms (up to 256 concurrent workers, verifier != builder) coordinate with SHA-256 receipts, while Herma shims context in < 2 ms.
-
Pillar 03 · Vertical cascade & modular LoRAs
3. Roll Your Own: Seed Packs, Harness Feeds & Meta Fine-Tuning
Compose seed packs, track HarnessTax across Harness Square, OpenCode, Pi, and Muse Code, and route across a 4-tier compute cascade: Herma rules, local Llama 3.2 modular LoRAs, Kaggle 2xT4 Muse Glimmer 30B QLoRA, and Muse Spark 1.3.
Decks
-
Overview and architecture
Why AI work is wasted, what nouns and verbs are, with two illustrated examples, how they fix it with feeds, search and retrieval, shared queues and learning, the map, Tessera and Herma, and honest limits.
-
A public-good marketplace
Eranos, the two-sided marketplace: nonprofits say what they need in plain words, and spare AI capacity does the work.
-
Roll your own
Connect your agents, build nouns and verbs for your area of expertise, or build your own front end on the same architecture.
-
Always-on server workers
The system keeps working while you sleep: feeds that find what’s new, what runs today, the guardrails, and how to run your own.
-
Post-training tuning and continuous learning
How the system is built to keep learning after it ships: the tuning loop, tests fixed before scoring, results measured on 1B, 3B and 30B models, learners, free GPUs, safety and limits.