intent-inference

intent-inference

Infers user intent from navigation, murmur, PLATO, and deliberation signals.

While constraint-inference learns constraint boundaries from user overrides, this service learns what the user is trying to do from their behavior patterns.

What It Does

The intent inference engine observes multiple signal sources and maintains a productive lane — a model of the user's current goals, priorities, and confidence levels.

Navigation → Where are they clicking/visiting?
Murmur     → What are they saying in fleet chat?
PLATO      → What rooms are they reading/writing?
Deliberation → What decisions are they participating in?
     ↓
  Productive Lane
  - primary_goals: [...]
  - confidence: 0.87
  - evidence: [...]
     ↓
  Fleet Bridge → tell agents what to work on

Architecture

src/
├── observers/
│   ├── navigation_observer.ts    — Tracks page/room visits
│   ├── murmur_observer.ts        — Monitors fleet chat signals
│   ├── plato_observer.ts         — Tracks PLATO room activity
│   └── deliberation_observer.ts  — Tracks decision participation
├── models/
│   ├── productive_lane.ts        — Goal/confidence/evidence model
│   └── intent_signal.ts          — Signal type + strength
├── inferrer.ts                   — Core inference engine
├── fleet_bridge.ts               — Push goals to fleet agents
├── storage.ts                    — Persist lane state
└── index.ts                      — Main loop (1-minute poll)

Productive Lane

The core data structure — a living model of what the user cares about right now:

{
  confidence: 0.87,
  primary_goals: ["ship dodecet-encoder", "fix fleet services"],
  evidence: [
    { signal: "navigation", strength: 0.9, topic: "dodecet-encoder", ts: "..." },
    { signal: "murmur", strength: 0.7, topic: "fleet services down", ts: "..." },
    // ...last 100 signals
  ]
}

Fleet Bridge

When the lane updates, the fleet bridge tells agents what to prioritize. This is how the fleet knows what the human wants without being asked explicitly.

Why This Matters

The fleet has 9 agents. Without intent inference, they'd all do what they think is best. With intent inference, they align around what the human actually cares about right now.

This is anticipatory alignment — the fleet starts working on what you need before you ask.

Usage

npm install
npm start

Polls every 60 seconds. Merges new signals into the lane. Pushes updates to fleet.

Ecosystem

• constraint-inference — Complementary: learns constraints from overrides
• flux-lucid — Intent vectors and navigation metaphors
• fleet-murmur — Signal source (fleet chat)
• dodecet-encoder — Consumer: lighthouse uses intent for task routing

License

MIT