Measure the cost of AI. Automate what's proven.

The Question DIAL Answers

For this specific task, at this specific step: can AI do it? Which model does it best? What does it cost? And how often does a human need to check?

DIAL tracks per-specialist, per-decision economics so you can answer these questions with data, not guesswork.

Per-specialist, per-decision:

$0.003costUSD

~200mslatencyMsec

94.2%alignment

Every proposal records cost, latency, tokens, and alignment against human ground truth

What DIAL Gives You

Exact numbers on whether AI can handle each step of your process — and what it costs.

Cost per Decision

Exact USD, token count, and latency for every proposal. Know what each AI specialist costs before you trust it.

Alignment Rate

Continuous measurement of how well each specialist predicts human choices. Not a training-time metric — a production metric.

Progressive Delegation

Proven decisions get faster and cheaper. Hard decisions stay with humans. The system tells you which is which, with data.

How It Works

Three steps from "humans do everything" to "AI handles what's proven."

Model

Define your decision process as a state machine. States, transitions, and prompts that describe what matters at each step.

Learn more →

Measure

Register AI and human specialists. DIAL runs decision cycles and measures which AI specialist best predicts what humans choose — tracking cost, latency, and alignment.

Learn more →

Delegate

As alignment is demonstrated, the system progressively delegates to the most trusted specialist. If alignment degrades, it reverts automatically.

Learn more →

Define. Register. Measure.

Define a state machine, register competing AI specialists, and DIAL measures which one best predicts human choices — with exact cost data.

app.ts

import { createSession, registerProposer, runSession } from "dialai";

const machine = {
  machineName: "content-review",
  initialState: "draft",
  goalState: "published",
  states: {
    draft: {
      prompt: "Review the content. Submit for review or keep editing?",
      transitions: { submit: "review" },
    },
    review: {
      prompt: "Check correctness, clarity, and tone. Approve or reject?",
      transitions: { approve: "published", reject: "draft" },
    },
    published: {},
  },
};

// Register AI specialists — each is a model + strategy + prompt
registerProposer({
  specialistId: "ai-reviewer-gpt4o-mini",
  machineName: "content-review",
  modelId: "openai/gpt-4o-mini",      // $0.003 per decision
});

registerProposer({
  specialistId: "ai-reviewer-llama",
  machineName: "content-review",
  modelId: "meta-llama/llama-3-8b",   // $0.0002 per decision
});

// Run to completion — DIAL measures alignment + cost
const session = await runSession(machine);