Decatron.aiGERVIS · EBM for AGI
EBM for AGI

EBM for AGI.Next-generation AI beyond prediction.

GERVIS replaces brute-force computation with energy-driven convergence — creating a deterministic, software-only foundation for operational cognition, world models and physical AI.

Explore the technology Start a technical discussion
DecaCell convergence core
Core DecaCell
Runtime GERVIS
Output Deterministic
34US provisional patent filings
Up to 99.5%measured energy reduction*
0observed output deviation in No-FP trials*
Software-onlygeneral-purpose hardware
DecaCell · The core

A new category of computation.

DecaCell is a post-transformer, energy-based compute core. It models constraints, relationships and tension dynamics, then converges toward the most stable valid state.

Energy-driven convergence
Up to 2¹⁰⁰ possible states in one convergence
Reduced computational work
Deterministic execution
Exact computation beyond fixed FP precision
The computational shift

Solutions converge instead of being exhaustively executed.

Conventional path

Predict and repeat

  • Clock-driven operation sequences
  • Brute-force tensor execution
  • Probabilistic next-token prediction
  • Heavy data movement
GERVIS path

Activate and converge

  • Energy-lawful resolution
  • Only the necessary work
  • Deterministic outcomes
  • Replayable execution
GERVIS · What makes it different

Five defining properties of a new intelligence foundation.

GERVIS combines a general energy-based solver with essence-based representations, learning without conventional training and closed-loop cognition — across tasks and industries.

02

Essences

Not word-based.
No tokens.
No weights.

01
GERVIS

General solver

Energy-based convergence

Constraints and relationships resolve through lawful energy dynamics toward a stable result.

03

Learns

Without training

New capabilities are taught, validated and retained without conventional model training.

05

Any task.
Any industry.

SciencePhysical AIHealthIndustry
04

Closed-loop cognition

1Observe2Reason3Plan / Imagine4Decide5Act6Learn7Improve
DeterministicReplayable outcomes
Up to 99% less energy*Lower compute cost
Real-timeAI that understands, decides and acts

* Internal measured results. Performance varies by workload, configuration and protocol.

What it enables

Real-time cognitive applications.

One deterministic foundation, designed for systems that must understand, decide and operate.

Agentic AI & digital workers

Persistent operational cognition for complex, multi-step work beyond one-shot prediction.

Physical AI & autonomy

Unified perception, planning and action for systems operating in the physical world.

World models & digital twins

Dynamic internal models that simulate possible futures and support real-time decisions.

Scientific discovery

Deterministic reasoning across large, constrained search spaces and exact mathematics.

Industrial operations

Continuous optimization across infrastructure, energy, production and complex systems.

Mission-critical intelligence

Replayable decisions, controlled execution and explainable operational behavior.

Measured foundation

Deterministic convergence across search, replay and exact mathematics.

The current benchmark suite tests DecaCell across constrained search, runtime stability, order independence and large exact-math execution.

* Internal measured results. Performance varies by workload, configuration and protocol. External validation is required for production claims.

TSP K=809.49 ms

9,702 nodes · 15,522 edges

No-FP runtime2.876 ms

average · one distinct output

Order trials0.000e+00

max deviation vs. baseline order

Exact math200k digits

integer square root executed in ~0.68 sec

The next foundation

Build intelligence that can operate — not only predict.

Talk with Decatron