Large Decision Models turn state, context and objectives into the next best action — one call at a time, across a full sequence of decisions.
Where an LLM predicts the next word, a Large Decision Model predicts the next action. It is trained on sequences of states, actions and outcomes so it can reason about consequences, trade-offs and long-horizon goals — the way an operator does, not the way a chatbot does.
Reads the full situation — inventory, prices, signals, constraints — as structured context.
Optimizes for the outcome of a whole sequence, not a single isolated step.
Returns the chosen action and the “why” — ranked alternatives and rationale.
Set the next price or discount given demand, competitors and margin targets — every hour, per SKU.
Decide what to reorder, route or hold when lead times and demand keep shifting.
Approve, flag or escalate a case with a consistent, auditable rationale on every call.
Allocate people, machines and jobs toward a target while constraints move in real time.
A single REST endpoint. Send the current state and the actions you allow — get back the recommended action, ranked alternatives, and the reasoning.
Pass your API key as a bearer token on every request. Keys are issued per environment.
# set your key export NEOINSTINCT_API_KEY="sk_live_..." curl https://api.neoinstinct.ai/v1/decisions \ -H "Authorization: Bearer $NEOINSTINCT_API_KEY"
POST the current state, the actions you permit, and the objective to optimize.
POST /v1/decisions { "model": "ldm-1", "objective": "maximize_margin", "state": { "sku": "A-1024", "stock": 240, "demand_7d": 512, "competitor_price": 19.90 }, "actions": ["hold", "discount_10", "raise_5"] }
You get the recommended action, a confidence score, ranked alternatives and the rationale to log or display.
200 OK { "action": "discount_10", "confidence": 0.87, "expected_margin": 3120.50, "alternatives": [ { "action": "hold", "expected_margin": 2870.00 }, { "action": "raise_5", "expected_margin": 2410.75 } ], "why": "Demand outpaces stock; a 10% cut clears units before competitor undercut." }