Current product boundary
The current Merion foundation is designed to prove the real-work-to-evaluation loop before expanding into hosted execution or training infrastructure.Available now
- Versioned contracts for work evidence, task candidates, evaluation tasks, experiments, and run records.
- Local SQLite evidence history and content-addressed artifacts.
- Guided factory initialization, source detection, readiness checks, capture, and outcome review.
- Deterministic task-formation gates with independent approvals and evidence audits.
- ATIF v1.7 import and Harbor task/result interoperability.
- A typed SDK and provider-neutral local runtime boundary.
- Deny-by-default cloud policy, manifest, receipt, resumable transport, and local sync-outbox contracts.
- Explicit metadata-only cloud configure, status, plan, queue, and disable commands with no automatic upload.
Not available yet
- Public package-registry installation.
- Hosted multi-tenant execution.
- A production authentication service, object-store transport, cloud worker, or actual network upload.
- Account, billing, or organization provisioning for the Intelligence Factory.
- Autonomous evaluation generation or approval.
- General-purpose model training or reinforcement-learning orchestration.
- Production customer-data ingestion.
- A broad experiment runner across hosted model and inference providers.