Energy AI agent · Trading and bidding
Wholesale Trading & Bidding AI Agent
Build bids in power and gas from your own positions and forecasts, record why each order exists, check it against your configured limits — and hold it for an authorised trader to release.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Trader release
- Pre-release
- After launch
- Agent Care
What this agent does
Builds the order, not the decision to send it
In
- 01
Positions, forecasts and market data from the customer's own ETRM, market-data and metering sources are ingested.
- 02
Instrument names, units and delivery periods are normalised, each carrying the source it came from.
Reason
- 03
Candidate bids and offers are constructed for the products and locations the desk has configured.
- 04
Desk limits — volume, location, product and price — are applied to every candidate order.
- 05
The reason the order exists is recorded against the positions and forecasts that produced it.
Decide
- 06
Position, forecast and limit inputs are read from the desk’s own book, never from another book.
- 07
Orders route to an authorised trader on the desk, who decides whether any of them is released.
Out
- 08
The order, its reason, its limit checks and the trader's decision are retained against the trade.
- 09
Write actions happen only inside the approval boundaries agreed during implementation.
Product statement
The agent proposes orders; an authorised trader decides whether any is released, and the firm remains the market participant of record.
Example workflow
One order, book to release
- 1Trading window opensDay-ahead, real-time or gas bidweek window on a configured market
- 2Book assembledPositions, forecasts, desk limits and market data, each with its named source
- 3Order proposedProduct, location, price, reason, confidence
- 4Controls appliedLimit checks, prohibited-pattern screens, cross-book review and confidence threshold
No human action required
Stages 1 to 4 run unaided and nothing reaches a market at any of them — the agent is building an order, and the trader's lane opens at the confidence gate.
Goes to an authorised trader to release.
Picks up a surveillance read first.
Trader release
The order is held with its reason, its limit checks and the confidence.
Authorise · Amend · Send to surveillance review
Authorised — sent to market
- 6Trading systems updatedOnly where write access and release policy allow it
- 7Outcome evaluatedAmend rate, screen outcomes, fill quality and post-trade surveillance findings
Amends
Every trader amendment, and every post-trade query, is counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Releasing an order into a power or gas market.
- Declaring a unit unavailable or off automatic control.
- Trading outside the limits configured for the desk.
- Using data from outside the customer's own book.
Automation boundary
Agent acts unaided
- ✓Build a bid from the customer's own positions and forecasts.
- ✓Record why each order exists, in terms a trader can defend.
- ✓Run the proposed order against the limits configured for the desk.
- ✓Hold the order, with its reason and its limit checks attached for the named owner.
Nothing is released ahead of a trader — exposure accrues per violation, per day, and disgorgement sits on top of it.
- Enabling a newly discovered trading strategy.
- A loss on one leg justified by another book.
- Orders whose only rationale is an allocation.
- Changes to limits, thresholds or release rules.
Example output
One order, annotated
Everything the agent proposes is attached to the book it was built from.
Bidding output · single order
Illustrative example- Market
- Day-ahead power
- Proposed order
- Offer at a single configured node, morning block
- Volume
- 25 MW
- Reason on file
- Own generation position
- Confidence
- 91%
- Limit check
- Inside desk volume and price limits
As received
Taken from the desk's own positions and forecasts — nothing on this side is written by the agent.
Book evidence used
- Owned unit position
- Own load forecast
- Desk limit set
Why this order exists
It sells a position the desk already holds and a person still decides.
What the score decides
Below the configured threshold the order picks up a surveillance read before a trader sees it.
Action
AuthoriseAmendSend to surveillance reviewValue
Where AI adds value
The same four claims, placed at the point in the workflow where each one applies.
Where the value lands
Value 01 – 04
Every proposed order
From the desk's own book
Build from the book
Draw only on the customer's own positions, forecasts and desk limits — pooled or competitor data is out of scope by construction.
A pattern is not a strategy
Routine bids and offers arrive already built, with their reason attached.
Send review where risk sits
Flagged patterns and low-confidence orders are marked, so the trader's read starts where exposure concentrates.
Build an evidence trail
The bid, the position and signals behind it and the trader who released it stay on the order.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Market systems
- PJM
- MISO
- CAISO
- ERCOT
- SPP
- ISO-NE
- NYISO
ETRM and scheduling
- Allegro
- Endur
- RightAngle
- Molecule
- Aspect
- Enuit
Market data
- Platts
- Argus
- ICE
- Weather and load forecasts
Agent
Trading & bidding
- Reads your book
- Builds the order
- Holds for release
Risk and surveillance
- Position and limit systems
- Trade surveillance tooling
Observability & evaluation
- OpenTelemetry
- Langfuse
- Supported monitoring/evaluation sources
Integration availability depends on the client's existing systems and API access.
Agent controls
Six layers between the model and the market
The controls are nested. What survives all of them appears in the map underneath.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Test whether an order may stand
- L3
- Puts the release in a trader's hands
- L4 – L5
- Show what the bid was built from
- L6
- Holds orders for the trader when signals degrade
How Nestack evaluates it
Evaluate the bidding workflow — not only the order that got filled.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the order the market sees
- E1Final-output evaluationDid the recorded reason actually explain the order?
- E2Step-level evaluationDid the agent use the right positions, forecasts and desk limits?
- E3Tool evaluationDid it read and write the correct book, market and product?
- E4Confidence calibrationDo low-confidence orders actually attract more trader amendments?
- E5Slice evaluationHow does performance change across specific markets and products?
- E6Business outcomeHow many orders needed a trader amendment or a post-trade correction?
Floor — the outcome the firm answers for
Failure modes
Where each failure originates in the agent
Seven ways an order goes wrong, set at its own stage.
Agent lifecycle
Direction of processing →
01 · Retrieval
1 mode
- MQ-03
Cross-tenant data leakage
A signal from outside the customer's own book reaches an order.
Stage gathers
Positions, forecasts, desk limits and market data
02 · Reasoning
2 modes
- MQ-04
Learned manipulation pattern
A profitable learned pattern matches a theory the regulator charges.
- MQ-06
Cross-book loss leg
A loss-making order is explicable only by another book's position.
Stage proposes
Product, location, price, the reason and confidence
03 · Tool / write
2 modes
- MQ-02
Price-driven unavailability
Availability is proposed as a function of what the award pays.
- MQ-05
Repeated machine footprint
Identical orders repeat into a pattern surveillance screens read.
Stage writes
Only where write access and approval policy allow it
04 · Output
1 mode
- MQ-01
Reason that will not hold
The recorded reason does not survive a trader's reading.
Stage returns
The order a trader releases and the market sees
05 · Change / Version
1 mode
- MQ-07
Silent limit regression
A model or limit change widens what the agent will propose.
Stage tracks
Model, prompt, limit sets and market config
- Sev-1 · an order leaves outside the boundary
- Sev-2 · a wrong order reaches a trader
- Sev-3 · inputs degrade, order routes to review
Affected slices
Overall quality can hide one bad book
Base rates differ sharply by market and by product, so a single desk-wide amendment rate says very little about where the rework actually sits. Nestack reports the trader-amendment rate by slice, not only in total.
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| Virtual and financial products | 9.6% | 3.3× | Review | |
| Cross-commodity positions | 8.1% | 2.8× | Review | |
| Newly configured markets | 4.9% | 1.7× | Watch | |
| Established power nodes | 2.6% | 0.9× | Normal |
Bar: trader-amendment-rate lift vs. established-node baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
The loop closes on a case, not a view
Explaining a bad fill does not close the cycle. A standing regression case does, and that suite is what the next order sent out is measured against.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Trader-amendment rate rises in a market slice.
- 02
Diagnose
If the cause is the inputs, the limits or the reasoning, orders are read with a trader until it narrows to one.
- 03
Improve
The fix is versioned against the orders that produced it.
- 04
Verify
The release stops until every affected case is passing.
- 05
Learn
The case stays in the suite, and the trading limits are amended.
Learn → Detect
The return edge. The next detection runs against a suite one case longer.
Typical build scope
Twelve workstreams across six weeks
The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, sources, bidding workflow, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Trading workflow discovery and boundary definition.Week 1
- 02Market, ETRM and data-source assessment.Week 1, Week 2
- 03Desk limit, prohibited-pattern and market mapping.Week 1, Week 2
- 04Position and market-data normalisation.Week 2
- 05Bid construction and reason binding.Week 2, Week 3
- 06Confidence scoring and flag routing.Week 3
- 07Trader release workflow.Week 3, Week 4
- 08Trading-platform and market integration.Week 5
- 09Intent and pattern cases.Week 4, Week 5
- 10Guardrails and release controls.Week 4
- 11Order-trail instrumentation.Week 5, Week 6
- 12Deployment, documentation and Agent Care handover.Week 6
12 workstreams · 6 weeks · bar shows the weeks a workstream is active — several run in parallel
Final scope and sequence confirmed in discovery
Engagement tiers
What each tier includes
Rows are the capabilities named in each tier's scope. Higher tiers include everything below them.
| Capability✓ in scope · — not at this tier | PilotOne market, one desk | ProductionProduction trading systems | AdvancedMultiple markets / entities |
|---|---|---|---|
| Introduced at Pilot | |||
| Bid construction from your book | ✓ | ✓ | ✓ |
| Trader authorisation | ✓ | ✓ | ✓ |
| Bid-quality baseline | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Reporting by market and product | — | ✓ | ✓ |
| Authorisation workflow in your systems | — | ✓ | ✓ |
| Write-back to your trading systems | — | ✓ | ✓ |
| Trading-platform integration | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-market and cross-product rules | — | — | ✓ |
| Multi-stage desk approvals | — | — | ✓ |
| High order volume | — | — | ✓ |
| Multi-market trading controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on integrations, workflow complexity, transaction volume, approval controls and deployment requirements.
- Separate from build
- Build pricing is separate from recurring Agent Care, which covers managed monitoring, evaluations, incidents and verified improvements after launch.
What we need from you
What you bring, and what we build with it
Each input maps to a piece of build scope and a week in the delivery timeline.
You bring
We build with it
01Your book structure and position sources
→Position and market-data ingestion and mapping
Week 102Representative historical orders
→Bid-construction baseline and reason binding
Week 203Your desk limits and prohibited patterns
→Desk limit, prohibited-pattern and market mapping
Week 104Access to relevant APIs, feeds or exports
→Market, ETRM and data-source assessment, then integration setup
Week 205Orders you would not want filled
→Pattern cases and the evaluation suite
Week 406Where an order must stop and wait for a trader
→Confidence scoring, flag routing, guardrails and release controls
Week 307Named authorised traders to release orders
→Trader release workflow, then pilot and production validation
Weeks 5–6
Nothing else is required
Deployment, documentation and Agent Care handover are ours.
Delivery timeline
Four phases across six weeks
Phases occupy the weeks the work really needs, so week 5 runs evaluation and launch side by side.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Trading workflow discovery, limit mapping and the automation boundary
W2Source integration and the bid-construction baseline
W3Bidding workflow, confidence logic and release controls
W4Evaluation suite, pattern screens and failure-mode testing
W5Trading-platform integration, pilot orders and targeted corrections
W6One trading period run under the desk, then handover
- Reading the band
- A bar spans only the weeks its work is named in. Week 5 carries two because that is how it runs.
- At the end of W6
- The period closes validation and Agent Care picks up the monitoring.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Energy AI agent
Build a trading agent around your desk's release chain.
Show us your book, your desk limits and who releases an order. Could you explain, months later, why each of yesterday's bids existed? That is the record we build and the boundary we set.
- Typical delivery
- 4–6 weeks
- Deployment
- Your desk and ETRM
- Trader release
- Pre-release
- After launch
- Agent Care available
Nestack Agents · Trading & bidding
AGT-EN-06 · Agent Care available after launch