From your knowledge to proven decisions
Rae is the latest development platform from Rainbird. It turns your policies and expertise into decision models, reasons over them deterministically in milliseconds, and proves every outcome with evidence.
Regulation · Policy · Expertise
The knowledge graph
Hundreds of rules · 11 reasoning layers · Every threshold a policy fact
The decision
Codify, reason, prove
One chain of custody from your policy to the decision your regulator reads.
- 1Codify
Captures expert knowledge as structured rules
RAEThe build platform that enables you to turn written policy and institutional knowledge into deterministic, auditable decision agentsNewRAELOG LANGUAGERaelog is the logic language that represents your policy and knowledge as a deterministic decision model.FOUNDRYRae Enterprise bolt-on that encodes your entire knowledge estate.Regulations, policies and expert interviews become a versioned Raelog model.
- 2Reason
Applies logic to reach verifiable conclusions
Raelog ReasonerThe Raelog Reasoner reasons over your models to ensure that identical facts produce identical decisions, each with a computed certainty and audit trail in milliseconds.Backward chainingA reasoning strategy that starts from the desired conclusion and works back to find only the data needed to prove it. asks only for the data it needs, then reasons symbolicallyUsing explicit, human-readable rules and logic rather than a statistical approximation. to an outcome.
- 3Prove
Delivers evidence for every decision made
The Evidence TreeThe Rae 'receipt' for every decision including the rules, facts and certainty that produced it.Every outcome ships with its rules, evidence and certainty, in natural language and JSONA plain-text data format, so the decision and its evidence can be read by any system without a special viewer..
Author, test and govern your own decision models
You own the models: human-readable, versioned and portable. Run in our cloud, a private cloud or on your own infrastructure.
Start from a use case accelerator
Open an accelerator in Rae, review the model and every test case, and adapt it to your sovereign policy.
Use Rae to build, test and publish your own models
For new greenfield projects, translate regulations, policies and expert knowledge into a structured Raelog model, with AI building and testing tools.
Simulation and testing
Build and run test cases against Raelog models before deployment. Validate edge cases and regulatory changes instantly.
Versioned like code
Every model is versioned in Rae, and every model is code: export it, hold it in your own repository, diff releases, and run it through your own CI/CD with its test suite.
Agentic knowledge engineering turns an existing documented procedure into a working Rae model in around one hour, ready for your experts to review.
The people who own the policy author and amend the model themselves, with no developers in the knowledge build process.
Every model published is available to your applications over the REST API and to your AI agents over MCP. Rae even helps you build the integration.
Deterministic reasoning at millisecond speed
Same case in, same answer out, every time, with the working shown.
Asks only for the data it needs
The reasoning engine doesn't evaluate every rule for every case. It starts from the question, works backwards to identify exactly which data is needed, and requests only that, minimising data exposure and processing overhead.
Judgement itself is always symbolic
The engine is a pure symbolic reasoner. It never approximates. Given the same inputs it always reaches the same conclusion, deterministic by construction, not by averaging across outputs.
Handles uncertainty without hiding it
Real-world decisions rarely have complete data. The engine reasons to the highest-confidence outcome supportable by available evidence, quantifies residual uncertainty, and flags what would change the result.
Confidence is calculated, never guessed
Every rule and every fact carries a certainty factor, and the engine combines them with a documented algebra. Each conclusion arrives with a computed confidence.
Data arrives by connection to your systems, by inference within the model, or by asking a person. The reasoning itself is symbolic end to end: no LLM in the decision loop, so decisions run at zero token cost
Every decision ships with its evidence tree
Every decision comes with an evidence tree which acts as its regulator-ready receipt. Each arrives with the full chain of rules, evidence and the certainty values that produced it, in natural language and JSON. Inspectable by humans, parseable by machines.
The checks
Reasoned · 100% certain
All seven assessments completed.
Because
- Income assessed and evidenced
- Expenditure assessed
- Affordability holds at the stressed rate
- Security and valuation assessed
- Credit history assessed: clean
- Deposit and source of funds assessed
- Term and retirement assessed
Policy fact
Acme lending policy requires all seven assessments to complete before any approval.
The record
Reasoned · 100% certain
The assessment record is complete.
Because
- Seven assessments were required by policy
- All seven have completed.
The candidates
Reasoned · 100% certain
Approve is a candidate decision.
Because
- The record is complete
- No policy rule is breached
- Nothing needs underwriter judgement
- No conditions are outstanding.
Policy fact
A decline outranks a referral, a referral outranks a conditional approval, and a clean approval ranks last: the strictest outcome always wins.
The decision
Reasoned · 100% certain
APP-3003 is approved.
Because
- Approve is the candidate at the strongest precedence
- Every check completed and none dissented.
351 rules · 37 relationships traversed · 4 data calls · 1 question asked · 0 LLM calls
Where Rae sits in your stack
A Rae model sits between your AI agents and your systems of record: the layer where judgement happens.
- Your agent needs a judgement it must not guess. It delegates.
- The Rae engine reasons deterministically over your encoded policy.
- The decision returns: certainty attached, evidence tree included.
REASONINGSAME FACTS
SAME DECISION
Portable models
You own the models: human-readable, versioned, portable. Run in our cloud, a private cloud or your infrastructure.
Published as an MCP tool
Publish any model as an MCP tool. Agents invoke it by name and get the decision back with its evidence tree.
Make the enterprise policy estate computable with Rainbird Foundry
Delivered as a programme with the Rainbird team.
Rainbird Foundry scales the same deterministic knowledge model architecture across the whole enterprise. It encodes regulation and policy estate-wide, resolving conflicts, overlaps and gaps. This knowledge layer is available to AI agents to ensure that all decisions are bound by regulation and policy.
Built for regulated industries
We carry high levels of certification and support multiple deployment options.
Security & compliance
- ISO 27001 · Certified
- HIPAA · Compliant
Deployment
- SaaS (multi-tenant)
- Private cloud
- Self-hosted
Integration
- REST API
- MCP, AI agent native
- Pre-built connectors
See Rainbird reason through your decision
Bring us a decision your organisation must get right. We will build a live Rainbird knowledge graph and run it for you in the room, on your data.