The underlying technology, grouped by purpose. A swimlane is a product surface, not an internal team.
Identity, Qubic, Compute, Gateway, Billing, ANN, Marketplace, Tissue, Dataset, Economy. The platform runs.
Training orchestrator, dataset connectors, retraining, A/B shadow deployment.
Trinary protocol v1. 8-ANN coordinator for material discovery. Phase 1 re-evaluation gate.
5 consensus policies. Signed IntentEnvelopes. 5 training recipes. Auto-publish to ANN.
AigarthPool (30/60/10), multi-sig treasury, pre-mainnet gates, hardware presale.
The Aigarth evolution proposal. Approved August 12.
Mutable genome, lineage, memory, fitness. Garden view + marketplace listing.
5 new tables, 13 routes, 4-tier compute, replication + challenge + reputation verification.
Workers that live in your data center, your laptop, or a partner's network. Cross-deployment reputation.
The Qubic on-chain processor reads Organism work items and commits 451-of-676 computor signatures.
The scheduler becomes an economic allocator, balancing cost, latency, and reputation.
The Neuraxon runtime as a backend for the Trinary envelope. Contingent on Neuraxon shipping.
The product form factor, from edge to cluster.
Edge developer device. Pre-orders open.
Workstation node for studios.
Rack-scale inference appliance.
Portable inference device.
Enterprise AI cluster.
Cross-region model parallelism.
What landed in Q3 2026, and what is on deck.
The Organism primitive and the Work Runtime. Three ADRs (Organism, Work Runtime, OC processor) plus 60+ new test cases. Twelve services in production.
Workers that run in your own environment, on a partner's network, or on a peer device. Cross-deployment reputation is the open question: how does a worker's standing in one network translate to another?
Hard things we are not pretending to have answered.
When do we move from replication + challenge + reputation to TEE or ZK? By volume, by value, by adversarial rate, or by a manual decision? The trigger changes the guarantee behind the "verified" label, so it matters.
When all three replicas disagree, who decides who is right? A deterministic re-run on a high-reputation worker, manual admin review, or a refund to the payer? Without a default, disputed work items never settle.