GIDE · Guaranteed Intelligent Dynamics Engine

Autonomous control you can prove is safe.

GIDE is a real-time control engine whose safety properties are formally verified against stated system and environment assumptions — one verified core, specialized at compile time, from quadrotors to live markets.

0.767 µs
CfC inference
13,000×
faster than real-time
49
theorems complete of 260 machine-checked
5
domains running end-to-end

// intelligence with guarantees

Verified, real-time, and self-improving.

Machine-checked safety

Control Barrier Functions verified as theorems in Lean 4 and Isabelle — over the mathematical model, not inferred from passing test runs.

Real-time

Closed-form continuous-time inference — the reference kernel measures 0.767 µs per forward pass, 13,000× inside a 100 Hz control budget.

Self-improving

Neuroevolution, plus a Self-Dev loop that writes and checks its own code — not yet fully autonomous.

// one engine, infinite domains

Specialized at compile time

Zig · C++ · Rust · Python · custom CUDA kernels

Cart-pole control
3D cart-pole balancing with machine-checked CBF safety proofs.
Quadrotor flight
6-DOF quadrotors with hybrid CfC / LQR / MPC control.
Financial markets
Live FX trading, a 36-objective curriculum, Oanda connector.
Language models
Entropy, grounding, and safety regulation for LLM control.
Autonomous software
A Self-Dev loop that specs, builds, and checks its own code — approaching its first fully autonomous cycle.

Running end-to-end means the domain builds against the same verified core and exercises it with its own harness and test suite — not a demo, and not a customer deployment. Maturity varies per domain and is stated on the evidence page.

Verified against stated assumptions.

Safety properties are stated as Control Barrier Function theorems and machine-checked in Lean 4 and Isabelle, over the mathematical model. Complete means mechanized end-to-end in both provers with zero local axioms; the remaining machine-checked theorems carry named local axioms, each documented and tracked for discharge. The assumptions they rest on, and what they do not cover, are published rather than implied.

Read the evidence →

Let's build something you can verify.

Safety-critical control, autonomous systems, formal methods — we'd like to hear from you.