Documentation · Get Started · Releases · Issues

Overview

DIAL provides a framework for answering a fundamental question: Given any task modeled as a state machine, how do you know — in dollars, time, and quality — exactly what it would cost to turn that task over to a minimally competent AI decision-maker?

DIAL starts from a deliberately pessimistic assumption: AI has no role. The default is that the task is too difficult for AI and only humans can navigate it. DIAL then provides the mechanism to prove otherwise, one decision at a time.

Key Principles

• Human Primacy: The human is always right — not because humans are infallible, but because humans have context that AI cannot access.
• Progressive Collapse: Over repeated decision cycles, measuring how well AI predicts human choices causes the multi-agent deliberation structure to progressively collapse into deterministic execution.
• Empirical Trust: Trust is earned through demonstrated alignment with human decisions, not assumed.

Install

npm install dialai

Quick Start

import { createSession, registerSpecialist } from "dialai";

// Create a session with a state machine
const session = createSession({
  machineName: "my-task",
  initialState: "idle",
});

// Register an AI specialist
await registerSpecialist({
  specialistId: "specialist.my-task.proposer.gpt-4",
  machineName: "my-task",
  specialistRole: "proposer",
  modelId: "gpt-4",
});

Packages

| Package | Description | |---|---| | dialai | Core library — engine, types, in-memory store, CLI, MCP server | | dialai/store-postgres | PostgreSQL store implementation (Kysely) | | dialai/migrations | Database migration runner for PostgreSQL |

Documentation

Full documentation is available at https://eloquentanalytics.github.io/dialai/.

Development

npm install          # Install dependencies
npm test             # Run tests
npm run build        # Build
npm run typecheck    # Type check
npm run lint         # Lint
npm run ci           # Full CI pipeline

Contributing

See CONTRIBUTING.md for development workflow and guidelines.

License

MIT