Grounding & memory infrastructure
GQueries
Persistent memory, authorized retrieval and fail-closed grounding for production AI systems.
Explore GQueries ↗Independent AI research & products
Research difficult questions.
Build verifiable systems.
AletheionAGI investigates state, memory, causality, grounding and intent—and turns bounded findings into infrastructure that can be inspected, tested and challenged.
Florianópolis · BrazilResearch-ledEvidence-bounded claims
The organization
One research program.
Independent product boundaries.
AletheionAGI is an independent Brazilian AI research and engineering organization. We build experiments to learn which mechanisms work, which fail and where their limits begin.
Our products share research foundations, but keep credentials, billing, data authority and operational responsibility separate. Customers may adopt either product without adopting the other.
Products
Research translated into
clear infrastructure boundaries.
Structured intent infrastructure
IntentParse
Raw user language becomes validated intent state with facts, constraints, preferences, unresolved dimensions and provenance.
Explore IntentParse ↗Embedded experience infrastructure
Widport
Intent, evidence and policy become a governed experience embedded directly in websites and applications.
Explore Widport ↗Open research project
ASM 1B
A staged effort to build, train and evaluate a Brazilian Aletheion State Model with approximately 1 billion parameters.
Support ASM 1B ↗Research agenda
Questions before conclusions.
Research lines remain hypotheses until a frozen evaluation supports a narrower claim.
State
How can compact persistent state preserve useful history without repeatedly exposing the entire context?
Causality
Which mechanisms improve temporal continuity, and under which measured conditions do they fail?
Grounding
How can retrieval, authorization and delivery controls prevent unsupported or cross-boundary claims?
Intent
How can systems distinguish explicit facts, bounded inference and unresolved intent before acting?
Method
Open evidence.
Explicit limitations.
- 01
Evidence before narrative
Claims identify their protocol, denominator, version and limitation.
- 02
Authority before autonomy
A model may propose; authenticated policy and canonical state decide.
- 03
Abstention is a result
Insufficient evidence should expose uncertainty instead of manufacturing certainty.
Work with us
Bring a question worth measuring.
For research collaboration, product partnerships, press or institutional matters, contact the organization directly.