The reasoning API for
AI engineers

Ship AI models with built in reasoning

Read the API reference

What it does

One API call. Predictions upgrade to actionable reasoning.

Embed the reasoning API beside a tabular AI model and call it like any inference endpoint. The Reasoning Engine returns inference outputs enriched with interpretation, patterns, counterfactual scenarios.

REST API and Python SDK.

Reason over OuterProduct models or any third-party AI model.

Reasoning runs real-time at inference. Your production model stays untouched.

import outerproduct as op

# Wrap any tabular model (ours or third-party)

model = op.get_model("risk-v4")

reasoning_model = op.reasoning.fit(data, model)

# Real-time explanations at inference

pred, exp = reasoning_model.predict_and_explain(record)

# Counterfactual scenarios

counterfactual = reasoning_model.scenario(record, target)

# Identify trends across data

op.reasoning.pattern_tracker.fit(reasoning_model, data)

import outerproduct as op

# Wrap any tabular model (ours or third-party)

model = op.get_model("risk-v4")

reasoning_model = op.reasoning.fit(data, model)

# Real-time explanations at inference

pred, exp = reasoning_model.predict_and_explain(record)

# Counterfactual scenarios

counterfactual = reasoning_model.scenario(record, target)

# Identify trends across data

op.reasoning.pattern_tracker.fit(reasoning_model, data)

import outerproduct as op

# Wrap any tabular model (ours or third-party)

model = op.get_model("risk-v4")

reasoning_model = op.reasoning.fit(data, model)

# Real-time explanations at inference

pred, exp = reasoning_model.predict_and_explain(record)

# Counterfactual scenarios

counterfactual = reasoning_model.scenario(record, target)

# Identify trends across data

op.reasoning.pattern_tracker.fit(reasoning_model, data)

Reasoning-as-context for agents

Every prediction enriched with reasoning

For every model score or classification, the Reasoning Engine outputs the reasoning an agent needs to make a defensible decision.

Tabular Interpretability

Identify the input signals that drive tabular model predictions per sample.

e.g. globally, counterparty risk and velocity dominate fraud loss; for this claim, a 10× spike + new merchant.

Counterfactual scenarios

The minimal change in input attributes that would flip model decisions.

e.g. email outreach → projected engagement increases 10%.

Decisioning Patterns

Surface emerging patterns that identify anomalous cohorts.

e.g. 1,258 new applicants with high risk scores were auto-denied due to thin file credit history.

Temporal Interpretability

Identify temporal signals that drive predictions across time-series models.

e.g. customer is 3x more likely to purchase a product because of their response to previous marketing campaigns.

Model experimentation

Tune model and policy versions on data before shipping.

e.g. policy v4 vs v3 across 48,902 scenarios → +2.1% in approvals and −0.4% in loss.

Feature utilization

Identify the data and features AI models actually use.

e.g. the model leans on 3 sources; one source is stale and under-covered.

What this means for agents

Reasoning powers agents you can trust

AI reasoning apps and agents are custom and context-specific. The App Factory gives business users the complete toolkit to compose apps and agents that automate decisions.

Traceable

Every action carries its reasoning (explanations, patterns, and scenarios grounded in data) making it defensible to a teammate, an auditor, or a regulator.