@superinstance/polyformalism-a2a

9-channel polyglot communication framework for multi-agent alignment.

Zero dependencies. ESM-only. Works in Node.js and browsers.

Install

npm install @superinstance/polyformalism-a2a

Quick Start

import { IntentVector, Channels, checkAlignment, selectFitting } from '@superinstance/polyformalism-a2a';

// Create intent vectors
const sender = IntentVector.zero()
  .set(Channels.Stakes, 0.9)
  .set(Channels.Process, 0.8);

const receiver = IntentVector.zero()
  .set(Channels.Stakes, 0.85);

// Check alignment
const report = checkAlignment(sender, receiver);
console.log(report.isSafe); // true
console.log(report.cosineSimilarity); // 0.998

// Select hydraulic fitting (right-size tooling to pressure)
const fitting = selectFitting(0.9);
console.log(fitting.name); // "DeepSeaSeal"

The 9 Channels

| # | Channel | Question | |---|---------|----------| | C1 | Boundary | What are we talking about? | | C2 | Pattern | How do pieces connect? | | C3 | Process | What's happening over time? | | C4 | Knowledge | How sure am I? | | C5 | Social | Who cares and why? | | C6 | Deep Structure | What's really being said? | | C7 | Instrument | What tools are available? | | C8 | Paradigm | What model of thought? | | C9 | Stakes | What matters vs what doesn't? |

API

Core

• IntentVector — 9D intent profile with salience + tolerance (backed by Float64Array)
• checkAlignment(sender, receiver) — alignment report with cosine similarity, draft margin, and per-channel warnings
• checkDraft(sender, capacity, speedFactor) — draft safety check with dynamic amplification
• toleranceStack(profile) — RSS tolerance: ε_total = √(ε₁² + ε₂² + ... + ε₉²)
• selectFitting(stakes) — hydraulic fitting selection (HoseClamp → DeepSeaSeal)

LLM Encoder

import { LLMEncoder } from '@superinstance/polyformalism-a2a';

const encoder = new LLMEncoder({
  endpoint: 'https://api.deepinfra.com/v1/openai',
  apiKey: 'your-key',
  model: 'ByteDance/Seed-2.0-mini',
});

const profile = await encoder.encode('The submarine hull pressure is critical');

Model-driven intent extraction via any OpenAI-compatible API. Replaces heuristic keyword matching with structured 9-channel extraction.

Intent Compilation

import { classifyPrecision, batchClassify } from '@superinstance/polyformalism-a2a/intent-compile';

Intent-directed constraint compilation — classify precision (INT8/INT16/INT32/DUAL) from C9 stakes channel.

License

Apache-2.0