CHAI
Platform · Intelligence Layer

CHAITHE INTELLIGENCE
THAT EXPLAINS ITSELF.

Nine engines producing intelligence is only valuable if that intelligence can be understood. Chai is the natural language layer that makes every output from every engine accessible to every stakeholder — from the dispatch desk to the boardroom.

Reads all nine engines Causal narration Counterfactual queries All plans · All tiers
CHAI · Intelligence LayerIllustrative example

What was the primary cause of the underperformance at Raven Ridge in February?

CHAI · querying KAIROS · HELIOS · FENRIR

Chai names the largest driver in plain language. Here, most of the month's revenue shortfall traces back to a wind forecast error — a sustained low-wind event the weather ensemble underestimated — and Chai reports KAIROS's estimated dollar effect together with its confidence range, not a single point number.

It then splits the rest of the variance across the other contributing causes — a regime shift SPECTER had not yet adapted to, and slippage on a CONSUL overnight bid — ranking them by how much each explains.

Finally it distinguishes the miss from the downside: the FENRIR worst-case floor still held, so the shortfall was in the central estimate, not the protected floor. Every figure Chai quotes is pulled live from the engine that produced it.

KAIROS HELIOS FENRIR SPECTER CONSUL
Capabilities

CHAI UNDERSTANDS
YOUR ENTIRE OPERATION.

Chai isn't a general-purpose chatbot with context about your site. It's an intelligence layer with direct read access to every engine output — current forecast, live regime state, simulation results, causal attribution, dispatch history.

01
Causal attribution in plain language

Chai draws on KAIROS's do-calculus model to explain revenue variance in narrative form. Not "these factors were correlated." Causally: "this is what caused it, with this confidence, and here's the counterfactual."

02
Live forecast narration

Ask "what does HELIOS see for the next 12 hours?" and receive a coherent narrative of the forecast, including the conformal prediction intervals, current regime state from SPECTER, and how FENRIR's adversarial scenarios map onto it.

03
Simulation result explanation

After STRIX runs a simulation, ask Chai to explain the P10/P50/P90 distribution, what's driving the spread, and what the Shapley attribution says about which aspects of the policy are contributing most to variance.

04
Counterfactual queries

Chai exposes KAIROS's counterfactual interface in natural language. "What would revenue have been if we had dispatched 20 MW more at 14:00?" — Chai constructs the do-calculus intervention, runs it, and returns the result with confidence intervals.

05
Stakeholder-appropriate output

Ask the same question as an operator and as a CFO — Chai detects the appropriate level of technical detail and adjusts. Operations gets interval-level precision. Finance gets monthly P&L impact and risk-adjusted projections.

06
CONSUL audit narration

Every autonomous bid decision CONSUL makes is logged. Ask Chai to narrate any session — why the initial bid was structured as it was, why a counter-bid was issued, what the market was doing at each round. Full narrative audit on demand.

Example conversations

WHAT YOU CAN ASK.

Regime & forecastIllustrative example
What regime are we in right now, and how is HELIOS reading the next 8 hours?

Chai names the current grid regime from SPECTER in plain terms — for example a high-solar, wind-suppressed pattern — and tells you when it began and what that kind of regime has historically meant for afternoon price volatility.

It then walks through HELIOS's forecast for the window: the central price path and the calibrated range around it, plus which of FENRIR's stress scenarios target that window — so you see both the expected case and the tail risks, in words.

SPECTERHELIOSFENRIR
BTM demand analysisIllustrative example
How close are we to hitting the demand threshold at Ironwood this week?

Chai reads PHALANX's billing-period trajectory and points to the specific window most at risk of setting a new demand peak — the day, the hours, and how likely a breach is if nothing is dispatched.

It tells you where your battery stands and whether dispatch is already pre-positioned to cover that window — so you know, in plain terms, whether the peak is handled or needs a decision.

PHALANXTARSIS
Revenue attributionIllustrative example
Show me the monthly attribution breakdown for January.

Chai uses KAIROS to break the month's revenue into its sources — how much came from energy arbitrage, how much from CONSUL's ancillary bidding, how much from SYNDEX managing market impact — each as a share of the total, in words rather than a raw table.

It then explains the change versus the prior month causally: which events added revenue, which subtracted it, and how confident the attribution is — so you can see not just what changed but why.

KAIROSCONSULSYNDEXSPECTER
Policy simulationIllustrative example
If we switched to the risk-aware RL policy, what would P10 revenue look like?

Chai has STRIX simulate both policies over many paths and then explains the difference in plain terms: a risk-aware policy typically lifts the downside case (the P10) at the cost of some upside (the P90), while an expected-value policy does the reverse.

Rather than pushing one answer, it frames the trade-off against your goal — protecting a revenue floor points one way, maximising expected value points the other — and shows which policy wins in more of the simulated scenarios.

STRIXHELIOS
Technical architecture

HOW CHAI IS BUILT.

Chai is powered by Claude (Anthropic) with a real-time context window that includes your current engine state. It doesn't rely on training data about your operation — it reads it live from every engine at query time.

Every Chai response is grounded in structured outputs from the engines — not interpolation. When Chai attributes a share of revenue variance to weather forecast error, that figure comes directly from the KAIROS attribution API for your site. The narration is generated; the data is read live from the engines.

✓
Live engine state access

Chai reads current outputs from all active engines at query time. Forecast, regime, attribution, simulation — all current.

✓
Structured grounding

Every numerical claim in Chai's response is grounded in a structured API response from a specific engine. No hallucinated numbers.

✓
Source attribution

Every response shows which engines were queried. Click any engine tag to see the raw output that grounded the response.

✓
All plans · no limits

Chai is included in every plan including the free 90-day pilot. No query limits. Full engine access from day one.

ENGINES CHAI CAN QUERY
FENRIR
Revenue floor · worst-case scenarios
SYNDEX
Market impact · bid ladder
KAIROS
Variance attribution · counterfactuals
SPECTER
Grid regime · personality embedding
CONSUL
Bid history · audit narration
HELIOS
Current forecast · conformal bounds
STRIX
Simulation results · Shapley attribution
PHALANX
Demand trajectory · peak risk
TARSIS
Savings breakdown · tariff analysis

ASK IT ANYTHING.
FREE FOR 90 DAYS.

Chai comes with every plan, including the free pilot. No query limits. All nine engines.