Phases
Phase 1
Working product
App, API, open models, bootstrap GPU supply, token, staking vault.
Users can send requests and receive streamed responses reliably.
Phase 2
External supply
Worker client, worker registry, marketplace listing, route weighting, provider earnings.
External GPUs complete production jobs and receive rewards.
Phase 3
Trust and verification
Spot checks, worker bonding, slashing, reputation, sharding research.
Invalid workers can be detected and penalized.
Phase 1 scope
The first build stays narrow. It produces a working inference product and enough internal architecture to support the worker marketplace.
In scope
- Chat interface and API with streaming responses.
- Model serving on reliable bootstrap GPU supply.
- Orchestrator with queues, model routing, and basic usage accounting.
- Staking vault configuration using existing infrastructure.
- Native worker spike that accepts one job and streams tokens back.
Out of scope
- Trustless verification for all production jobs.
- Large model sharding across multiple consumer GPUs.
- Slashing and dispute resolution.
- Full marketplace replacement of bootstrap supply.