Unified ASI Framework
29 cognitive modules, a shared Cognitive Blackboard, and a live ZK Rings↔Ethereum bridge. The cognitive backbone of FETCH-AGI.
Every cognitive capability is encapsulated in its own module — composable, testable, independently deployable. They communicate through the shared Cognitive Blackboard, achieving collective intelligence without a central controller.
Showing 12 of 29 modules · Full list in source repository
Every module publishes findings. Every module reads. No central controller — emergent coordination through shared state.
The Cognitive Blackboard is borrowed from classical AI architecture but evolved for the 21st century. It provides a lock-free shared address space where every cognitive module can post findings without waiting for a controller to dispatch them. Modules observe each other's work directly, enabling emergent cognitive resonance — where discoveries in one domain spontaneously activate related processing in another.
Under the hood: a lock-free MPMC ring buffer, versioned fact trees with confidence decay, and a publish/subscribe topology that keeps latency below 1 millisecond for 99th-percentile reads even with all 29 modules writing simultaneously.
Zero-knowledge proofs meet Ethereum settlement. Agents prove what they know without revealing it — and get paid for being right.
Agents generate zero-knowledge proofs of their knowledge claims using BLS12-381 Groth16 circuits. Proofs are verified on-chain via Ethereum smart contracts on Sepolia testnet. Agent-to-agent payments settle trustlessly — no intermediary, no trust required. Cognitive merit becomes economic signal.
Full ZK proof system over the BLS12-381 pairing-friendly curve. Knowledge claims provable without revealing content. Succinct proofs verified in O(1) on-chain. Trusted setup with multi-party ceremony.
Agents earn and spend ETH. Meritocratic compensation for verified cognitive contributions. Payments routed through ring topology — no single point of trust. Supports microtransactions at sub-cent granularity.
Distributed across node0–node5. Each node independently verifiable. Byzantine fault-tolerant ring consensus. Schnorr threshold signatures for collective finality. No single node can censor or forge.
Four layers from cognition to consensus. Each layer speaks to the next through a defined protocol — composable, independently upgradeable, formally verifiable.
asi-build is the cognitive backbone. These systems complete the AGI architecture — memory, training, and orchestration working in concert.
Active-inference Training with Learned Adaptive Stigmergy. Custom Rust inference engine for large language models. Champagnat n-morphic framework. A100 GPU benchmarks: 4.1 tok/s OLMo-3-7B, 37.7 tok/s SmolLM2-135M. 528 tests.
Stigmergic memory palace engine. Embedded graph database with pheromone trails. Semantic A* pathfinding. 13 Rust crates, 28 MCP tools, 720+ tests. Stores and navigates all cognitive knowledge.
Multi-agent coordinator fusing asi-build and GraphPalace into a living AGI architecture. 5-agent team with scheduled intelligence cycles. Owns the Cognitive Blackboard. Bridges research to deployment.