GiP: Neural Computation Protocol (NCP) v0.3.2 – Deterministic WASM Bricks + YAML Graphs for Fast, Auditable Agentic Systems

[madeinplutofabio]

Title:
GiP: Neural Computation Protocol (NCP) v0.3.2 – Deterministic WASM Bricks + YAML Graphs for Fast, Auditable Agentic Systems

Author: madeinplutofabio
Date: April 20, 2026

Summary

NCP provides a lightweight, open protocol for composing tiny sandboxed WASM “Bricks” into static YAML graphs. It routes 90–97 % of agentic work (routing, validation, classification, policy checks) to deterministic microsecond paths, escalating to any LLM (or decentralized model) only when truly needed.

Problem

Current agent frameworks still push even simple deterministic logic through expensive LLM loops. This explodes cost and latency in high-volume pipelines and creates audit / prompt-injection risks in decentralized setups.

Proposed Solution – NCP

• Bricks: Pure-functional WASM modules (no FS, no network, no ambient authority). Deterministic by design.
• Graphs: Static YAML files defining directed graphs with typed edges, routing policies, threshold gating, and field mapping.
• Runtime: Reference implementation in Rust + Wasmtime. Enforces hard limits and produces full replayable traces with hashes + provenance.
• Hybrid model: Bricks can emit confidence scores; the graph decides deterministically whether to take the fast path or escalate.

Current Status (as of Apr 20, 2026)

• Latest release: v0.3.2 (Apr 14, 2026)
• Spec: v0.2.3 (stable) with full JSON Schemas
• Benchmarks: 15–34 µs deterministic path; 10–33× latency/cost reduction in 90/10 and 97/3 hybrids (see BENCHMARK.md)
• Phase 1 & 2 complete, Phase 3 (LangGraph / MCP integrations) in progress
• License: Apache-2.0
• Repo: https://github.com/madeinplutofabio/neural-computation-protocol

Why it fits Gonka / decentralized infra

NCP is local-first, sandboxed, and trust-minimized. It dramatically reduces reliance on expensive/decentralized inference while keeping everything auditable and replayable.

Looking forward to community feedback!