Know exactlywhen to stop trusting your AI— before it costs you.

Rebound is middleware that sits between your AI/ML systems and the decisions they drive. It computes the Lyapunov-bounded predictability horizon k* for your data stream, maintains a Rashomon set of equivalently-accurate models, and routes every request to the model that minimizes domain-specific risk at the requested horizon — or refuses the request when k > k*.

Experiment trackers (MLflow, W&B) record offline runs. Drift monitors (Fiddler, Arize) detect degradation after it has already occurred. AutoML platforms (DataRobot, H2O) optimize for aggregate validation accuracy. Rebound is the only system that selects models by decision utility at horizon k using Lyapunov-derived predictability bounds — and refuses to act past those bounds. There is no direct competitor today.

Four product surfaces ship on the same core: FinAlgo for quantitative trading and model risk, CodeSentinel for agentic coding agents (VS Code extension + GitHub Action), GridOps for renewable dispatch and grid scheduling, and TacticalHorizon for defense decision-support simulation. Each speaks the vocabulary of the buyer in that vertical.

Three tiers. Pulse is a hosted REST/gRPC API for startups and individual evaluators. Command is a multi-tenant SaaS dashboard for enterprise design partners with RBAC, SSO, and webhooks. Fortress is the air-gapped on-premise build for defense and classified workloads — FIPS 140-2/3, IL5/IL6 capable. Models stay in your environment; only metadata and decisions cross the boundary if you allow it.

The Compliance & Audit Engine is purpose-built for EU AI Act Article 9 (Risk Management), FINRA SR 11-7 (Model Risk Management), NIST AI RMF, DoD AI Ethics Principles, and FCA AI Governance. Every selection produces an immutable, signed audit record with the rationale, the model chosen, the rejected alternatives, and the k* at decision time. Conformity reports export with one click.

Anything that exports to ONNX, plus first-class support for PyTorch, TensorFlow, JAX, scikit-learn, Hugging Face Inference Endpoints, AWS SageMaker, GCP Vertex AI, and Azure ML. Streaming inputs come in over Kafka or REST. The Python SDK is available for trial users today.