Persistent AI infrastructure that remembers, recovers, and proves its work.
Website | Documentation | Whitepaper | Ecosystem Map | Public Core
DreamNet is building a model-neutral operating layer for persistent agents: bounded missions, portable capabilities, recoverable state, explicit authority, independent verification, and content-addressed receipts.
| Repository | Purpose |
|---|---|
| Institutional Protocol | Run the five-minute guided journey from competency and bounded work to disputed claims, independent verification, federation, and scoped reputation |
| DreamNet Public Core | Typed contracts for assignments, Capsules, claims, receipts, and deterministic validation |
| DreamLoops | Persistent operating loops, succession, recovery, lineage, and portable capability Capsules |
| ToolGym | Agent exercises, deterministic grading, practical verification, and portable competency evidence |
| Claim Factory | Credentialed claim production with separate, independent verification factories |
| Cerberus | Offline supply-chain inspection for repositories, packages, hooks, workflows, and agent bundles |
| GitGrid | Git-native, append-only events, provenance, receipts, and rebuildable indexes |
Human goal
-> assignment
-> capability Capsule
-> bounded DreamLoop
-> authorized tools and models
-> execution
-> producing Claim Factory
-> independent Verification Factory
-> receipt and Proof Drop
-> durable graph and memory
The model can change. The evidence contract remains inspectable.
observation
-> evidence
-> receipt
-> claim
-> counterclaim
-> independent verification
-> portable knowledge
-> DreamLoop evolution
DreamNet is not optimized merely to produce answers. It is designed to turn execution into inspectable, portable knowledge that another agent, person, or independent organism can verify.
The production monorepo, customer data, credentials, internal prompts, routing policy, and private swarm topology remain private. Public repositories are scoped contracts, reference implementations, and runnable product surfaces.
The independent ZOL project has merged DreamLoops-based persistent-agent work, including its persistent-agent graft and Bonfire state adapter.
- Receipt-backed operational graphs
- Persistent agent recovery and succession
- Independent Claim Factories
- Supply-chain defense for agent workspaces
- ToolGym competency receipts
- Local-first memory and context transport
- Human-controlled market intelligence and simulation
DreamNet is early. The goal is not to claim perfect autonomy; it is to make useful autonomous work bounded, reproducible, recoverable, and verifiable.