Example recommendation · intradayExample

Sell 18 MW, 18:00–18:15 CET

Forecast P50 / P10
112.4 / 101.2 MW
Position after trade
100.0 MW
Expected imbalance
−0.6 MWh
Scenarios evaluated
2,000
Limit check
within max bid 110 MW
Rationale: sells the gap between the current position and P50 while keeping cover at P10.Needs approval
Example. Values are illustrative.

Dispatra · Decide

Turn a forecast range into a decision

Dispatra optimizes over thousands of forecast scenarios to choose bids and setpoints that hold up across the range, explains why, and acts only inside your limits.

Capabilities

What Dispatra does

Optimization

Decisions under uncertainty

Stochastic optimization over scenario trees, so a bid is chosen for the whole range, not just P50.

Markets

Day-ahead, intraday, balancing

Bid curves and nominations shaped for each market's rules and gate closures.

Rationale

A reason for every action

Each proposal states the forecast, constraint and trade-off behind it in plain language.

Guardrails

Limits outside the model

Volume caps, ramp limits and market rules are checked before anything is sent.

Autonomy ladder

From L0 to L3

You choose per asset and market how much Dispatra may do without a person.

Attribution

Decision to euros

Each decision is compared with the agreed baseline once settlement data arrives.

How it works

Inside Dispatra

  1. Scenarios

    Skycast quantiles and ensemble members become a scenario tree.

  2. Optimize

    cuOpt searches bids and setpoints across the tree within the market window.

  3. Check

    NeMo Guardrails and hard limits check market rules and operating bounds.

  4. Explain and act

    The agent writes the rationale; the action goes to an approver or executes within bounds.

Autonomy

You choose how far it goes

Each level is a setting per asset and per market. Pilots stay at L0.

L0Recommend

Dispatra proposes a bid or setpoint with its reason. A person decides.

Pilot default
L1One-click approve

The proposal is pre-filled; an approver confirms it before gate closure.

Planned
L2Auto within bounds

Acts alone inside hard limits; anything outside goes to a person.

Planned
L3Supervised autonomy

Runs a market or asset under supervision, with every action logged.

Later

Specification

Inputs, outputs and hardware

ItemDetail
InputsSkycast scenarios, positions, prices, market rules, asset limits
OutputsBids, nominations, battery and curtailment setpoints, rationales
AutonomyL0 recommend (pilot default); L1–L3 planned
AuditInputs, run, rationale, approver and outcome per decision, exportable
HardwareHGX or H100 for optimization; DGX for training the policy and agent
ServingTriton, TensorRT and NIM for low-latency inference

Questions

About Dispatra

Does Dispatra trade on its own?

Not in a pilot. Pilots run at L0, where Dispatra recommends and a person decides. Higher levels are settings you choose later, per asset and market.

How do you stop a bad decision?

Hard limits sit outside the model and are checked on every action. A recommendation that breaks a limit is blocked and flagged, not sent.

Can we see why it chose a bid?

Yes. Every proposal shows the forecast, the scenarios that drove it, the constraint that was binding and the expected outcome.

Which markets first?

European day-ahead and continuous intraday with 15-minute settlement. Balancing markets follow with design partners.

See Dispatra on your own sites

We start with a backtest on your history and show the result side by side with your current approach.