Skip to main content
NMP is a next-generation, high-performance binary transport mesh designed specifically for advanced Artificial Intelligence Agent communication. Conceived as the conceptual and technical evolution of existing context protocols (like the Model Context Protocol), NMP radically shifts the paradigm from pulling massive data chunks towards secure Logic-on-Origin (LoO) execution. LoO is the Postulate of Origin (Execution Core) of the neural mesh: the data is sacred and must never leave its physical location unless mathematically aggregated. If MCP was built as the “USB-C port for AI applications” to standardize connections, NMP is the fiber-optic nervous system that enables true decentralized intelligence.

​
Why NMP?

In the current landscape of autonomous agents, traditional client-server architectures face severe bottlenecks. Transferring gigabytes of raw data (logs, databases, repositories) over HTTP/JSON across the internet for an LLM to parse, reason, and filter is increasingly inefficient, slow, and expensive in terms of token context limits. NMP solves this by completely inverting the architecture: We move the math to the data, not the data to the math.

Logic-on-Origin

Instead of downloading giant datasets, Agents dynamically compile their reasoning step into a microscopic WebAssembly (.wasm) module and push it exactly where the data lives.

Multi-Layered Security (Tier-1)

Zero-Trust AST Sandboxing (WASI), Egress PII verification with Military-Grade algorithms (Luhn, NIST boundaries, Safe Words), and Post-Quantum cryptography.

Zero-Shot Autonomy

NMP Auto-educates remote LLMs on-the-fly. If an Agent violates the LoO paradigm, the Server cognitively instructs the Agent to rewrite its WASM payload securely.

Binary High-Performance

Built entirely on compiled Protobuf messages multiplexed over Tonic (gRPC). Say goodbye to slow JSON-string parsing blocking your event loops. Payload sizes drop by over 60%.

Multi-Core Scalability

The Node.js SDK leverages a Native Worker Pool (piscina), bypassing the single-threaded V8 limit to achieve transactional parities identical to massive Rust setups.

Mathematical Verification

Built-in architectures for Zero-Knowledge Proofs (ZK-SNARKs) allowing origin servers to Cryptographically sign verifiable receipts (ZK-Receipts) of the execution.

Decentralized DHT Mesh

Zero central authorities. Agent and Developer discovery happens instantly and securely over a global libp2p Kademlia Distributed Hash Table.

​
The Power of Logic-on-Origin (LoO): Innovative Use Cases

The Logic-on-Origin (LoO) paradigm is not just a performance optimization; it is our undeniable architectural postulate. It fundamentally unlocks completely new ways to process restricted data across global networks without violating privacy boundaries. By sending a verified mathematical algorithm to the data source rather than extracting the data to the algorithm, we eradicate Data Gravity entirely.

​
1. Healthcare & Bio-Informatics (Zero-Exfiltration)

An AI Doctor Agent wants to analyze 10 years of patient history across multiple hospitals to find correlations between specific medications and side effects. In traditional computing, this is impossible due to HIPAA and GDPR: hospitals cannot export thousands of medical records (containing Personally Identifiable Information). With NMP, the Agent injects a securely sandboxed WebAssembly filter. The hospital’s Data Node runs the logic locally against its database. The Agent receives only the final mathematical conclusion (e.g., “85% correlation found for Medication X with target side effect”) backed by a cryptography receipt, without ever seeing a single patient’s name.

​
2. Financial Auditing & High-Frequency Trading (HFT)

An AI Auditor is tasked with validating millions of transactions across international banks for fraud detection. The sheer size of a global banking ledger, combined with banking secrecy laws, makes transferring this data to a central cloud LLM illegal and unfeasible. Via NMP, the auditor injects a fraud-detection heuristic directly into the Bank’s secure Nitro Enclave. The logic runs over the raw ledger at bare-metal speeds, returning only the flagged anomalies.

​
3. Edge Computing & Massive IoT Telemetry

Smart factories, autonomous drones, or satellites generate gigabytes of telemetry per second. A central AI cannot ingest this sheer volume in real-time. Instead, NMP pushes dynamic intelligence logic down to the Edge antenna. The Agent installs a perpetually sleeping WASM Watchdog on the satellite. It processes terabytes of data locally in orbit, and asynchronously streams an alert via gRPC only when a critical failure pattern is detected.

​
4. Decentralized Collaborative AI Research

Multiple pharmaceutical consortiums want to train advanced models on their combined genomic databases without sharing their raw proprietary molecules with competitors. They use NMP to push training algorithms to each other’s origins, extracting only encrypted gradient updates to assemble the final shared model securely.

​
Who is NMP for?

NMP provides significant benefits across the entire AI ecosystem:
  • For AI Application Developers: Enable your agents to analyze remote massive repositories or databases instantly without waiting for megabytes of text to stream across the network.
  • For Data Owners & Enterprise: Expose your local tools and sensitive data securely. Since the Agent’s logic runs in a mathematical sandbox (WASI) on your machine, your raw proprietary data never leaves your local network. It only exports the final answer.
  • For the AI Community: A universal, open, and mathematically verifiable standard for deploying dynamic intelligence models and autonomous watchers.

​
Core Pillars of the Neural Mesh

  1. Capability-Based Security: Permissions in NMP are not binary strings. They are File Descriptors actively mapped into the Sandbox. An agent can only see what it was strictly granted.
  2. Post-Quantum Handshakes: Intent negotiations use Kyber (ML-KEM-768) ensuring “Harvest Now, Decrypt Later” state-level attacks are completely mitigated, encapsulating all data via Symmetric AES-256-GCM.
  3. Real-Time Watchdogs: Agents can push persistent .wasm modules that sleep on the server for free, and asynchronously wake up to stream an alert via gRPC only when an anomaly is detected.
  4. Hardware Blind Computing (TEEs): Future-proof architectural designs ready to execute the Wasmtime engine inside AWS Nitro Enclaves, assuring Memory-level encryption for High-Compliance environments.
  5. Zero-Time AST Guardians: The SDK preemptively inspects and destroys .wasm binary trees that attempt illegal namespace mappings before they even reach the initialization engine.

​
Getting Started

Ready to dive in and modernize your AI architecture? Choose your path:

Quickstart Guide

Launch your first NMP Server and Client in under 5 minutes using the official SDK.

Architecture Deep Dive

Understand how Kademlia, gRPC, and Wasmtime intertwine to create the Logic-on-Origin paradigm.