Rae
Live webinar: Thu 3 Dec, 3pm GMT
Rae: Reasoning for your code and your agents

Your AI agents are clever, but they shouldn't be making decisions.

Point Rae at regulation, policy and institutional knowledge, and it builds a deterministic model you can call over MCP or REST. Every answer comes back with its logical proof, and no language model sits in the decision path. Same facts, same answer, every time.

0.22 ms
An answer 32 levels deep, across 1.9 quadrillion derivation paths
60 ms
A complete 65-check case, on a real model from a global bank
100%
SAME ANSWER, IN THE SAME ORDER, ON 181,896 REPEATED RUNS
8,845 of 8,845
Answers whose evidence passed every independent proof check

See full benchmarks

Live demo · no account needed

Set the facts. Watch it reason. See why.

Below are examples of Rae models making decisions live. Set the facts however you like, and every answer is made deterministically with the evidence tree included.

This model identifies the best poker hand in any five cards. Deal a random hand or edit the inject payload directly, then click Decide. Rae will make a determination and provide the evidence tree behind the decision.

Input Your five cards

Output Answer and evidence

Ready to DecideReady
Ready to Decide
Why it holds up

System 2 reasoning in milliseconds. Every answer with its evidence.

Daniel Kahneman described two modes of thought. System 1 thinking is intuitive and probabilistic working largely from pattern recognition. System 2 thinking is deliberate and logical and works through the steps. Newer Systems 1 decision models return a fast, calibrated guess. These are a reflex, and do not come with a proof.

Complex policy decisions need System 2 thinking, a chain of inference worked through methodically with rules applied and each step auditable. Until now that has meant a chain of agents, often with a person in the loop, producing a result that is still non-deterministic, imprecise, and not guaranteed to adhere to policy. Rae automates System 2 reasoning in one deterministic step, in milliseconds, bounded by encoded policy with no language model in the reasoning.

In

A person or an event

A transfer arrives. An alert fires. A customer asks for help.

System 1

Co-pilot, agent or application

A language model handles language: reading, routing, replying.

System 2

Rae reasons over a knowledge encoding

Every relevant rule in the model applied to the facts, in milliseconds.

Out

Decision, certainty, evidence tree

The decision comes with a chain-of-reasoning so your agent can safely act.

Testable

Built, not trained

Knowledge becomes a computable asset which is encoded, refined and easily tested through the creation and running of evals.

Deterministic

Same in, same out

The same facts and the same version of a model always give the same decision and the same evidence.

Evidenced

The tree comes with it

Every answer includes its evidence tree, a proof describing how the decision was achieved including a snapshot of all data and rules.

Models

Rae models can be of any size and complexity, built and refined from multiple knowledge sources.

Rae enables you to build models that include an encoding of regulation, policy and institutional knowledge of any size and complexity. Rae then reasons deterministically over the model to automate System 2 reasoning in milliseconds, bounded by encoded policy with no language model in the path.

Rae reasons over a model you build and own.

Strong typology signalCandidate outcomeAdjudication outcomeContributing evidenceRationaleRecommended next step

Rae encodes knowledge as a mathematically complete symbolic model capable of representing facts, rules, certainty, weights and behaviours, but with no requirement to chain logic. Ask a query and inject available data, and the engine chains back from the goal and touches only the rules and data the question requires. It asks clarifying questions if it does not have what it needs or cannot infer it. Models are versioned when policy changes, and Rae includes comprehensive testing capabilities.

The same model can be a headless decision service when every fact is supplied, but an interactive advisor when some are left for a person to answer.

Code

Call it from anywhere: your agent, application or chatbot.

Add a decision agent to any process via an MCP endpoint or call it over REST. Rae provides tooling to help you build and manage your integration.

# 1 · Start a session
curl https://api.rainbird.ai/start/a665605a-a7c4-48a5-ad12-1b82d2600c43 \
  -H "X-API-Key: $RAINBIRD_API_KEY"
# → { "id": "<session id>" }   then: export SESSION_ID=<session id>

# 2 · Inject facts
curl -X POST https://api.rainbird.ai/$SESSION_ID/inject \
  -H "X-API-Key: $RAINBIRD_API_KEY" \
  -H "content-type: application/json" \
  -d '[
  { "subject": "HMKD Harbormark Dollar", "relationship": "has issuer permit type", "object": "Federal Qualified PPSI", "certainty": 100 },
  { "subject": "HMKD Harbormark Dollar", "relationship": "issuer domicile country", "object": "United States", "certainty": 100 },
  { "subject": "HMKD Harbormark Dollar", "relationship": "issuer regulating country", "object": "United States", "certainty": 100 },
  { "subject": "HMKD Harbormark Dollar", "relationship": "issuer holds us reserves", "object": true, "certainty": 100 }
]'

# 3 · Query
curl -X POST https://api.rainbird.ai/$SESSION_ID/query \
  -H "X-API-Key: $RAINBIRD_API_KEY" \
  -H "content-type: application/json" \
  -d '{ "subject": "HMKD Harbormark Dollar", "relationship": "reserve condition excused" }'
# → { "result": [ … ] } — or { "question": { … } }: answer it with /response

Start building decisions your agents can rely on.

Turn your regulation, policy and institutional knowledge into a decision model. Call it from your code or agent, and every answer comes with evidence.