Independent AI research organization
AletheionAGI
We investigate state, causality, memory, grounding and intent—and turn bounded findings into verifiable systems.
Mission
Turn difficult questions into verifiable systems.
AletheionAGI is an independent Brazilian AI research and engineering organization. We investigate whether state-based causal architectures can deliver useful memory, efficiency and measurable behavior at different scales.
Our goal is not to defend a predetermined conclusion. It is to build the instruments required to discover which mechanisms work, which fail and under what conditions.
How the work fits together
Theory, research and product are different layers.
That separation prevents a mathematical inspiration from becoming an automatic claim about an AI system.
Directional Relational Manifolds
A mathematical framework for variable effective dimension, rank transitions and path-dependent memory.
Aletheion State Models
Experimental research into causal models that compress history into persistent state and learn how it evolves.
ASM-CM · ASM-R · ASM-S · ASM-F · ASM-X
ASM-CM is the promoted compact-memory architecture; the remaining variants preserve relational transition, efficiency and geometry for comparison.
How we work
Operating principles.
Evidence before narrative
Hypotheses are separated from results, and every claim carries its protocol, scale and limitations.
Architecture before branding
Variants survive through experimental performance, not attachment to a name or theoretical origin.
Reproduction as a method
Code, configurations and evaluation criteria should support auditing and independent repetition.
Risk made explicit
Negative results, methodological uncertainty and technical blockers are part of the scientific record.
Independence
Open to scrutiny, collaboration and changes in direction.
Independence lets us explore ideas outside the consensus, while increasing our responsibility for transparency. Results must travel with methods, artifacts and limits.
We do not present experimental results as proof of general superiority. The ASM program remains willing to replace variants and remove mechanisms when evidence points elsewhere.
Collaboration
Want to evaluate or build with us?
We welcome conversations with engineers, researchers and organizations interested in grounded memory and reproducible systems.