Agentic Memory Needs More Than Retrieval
Persistent agent memory is state-transition machinery: evidence, projections, activation, context, procedures, authority, lifecycle, revalidation, and recovery must remain distinct and governed.
Open cognitive infrastructure
So they CLAIM...? We keep receipts.
Know where an answer came from. Rosetta connects claims to their sources, context, and history—so people and AI systems can reason with evidence that remains open to inspection.
SourceDocument01
A laboratory report records a sample temperature of 21 °C.
The original record anchors the chain. A receipt can identify the artifact, its author, and the time it was recorded.
Inside the receipt
Example record · LAB-001
ObservationMeasurement02
The recorded temperature is within the stated 20–22 °C range.
An observation describes what the record supports. It retains the measurement context and does not silently become a broader claim.
Inside the receipt
Derived from · LAB-001
InterpretationInference03
This reading is consistent with the stated storage condition.
An interpretation adds reasoning. Its assumptions and limitations belong beside the claim: one reading does not establish continuous compliance.
Inside the receipt
Assumption · Sensor reading is accurate
ProjectionForecast04
Further readings could help establish whether the condition persists.
A projection concerns what may follow. It is conditional, and later evidence can revise it without erasing the earlier reasoning.
Inside the receipt
Uncertainty · Future conditions are unknown
Illustrative example · Traceability is not proof of truth.
Research & perspectives
Questions worth getting right.
Evidence reviews and research proposals on auditable intelligence, technological change, and the institutions that must adapt.
Ontological Mixture of Concepts
OMoC treats reasoning as a semantically routed computation graph: task-local concept signatures select heterogeneous cognitive operators instead of static agent roles.
Cognitive Tapestries via Semantic Latticing
Semantic latticing treats context as a compiled, provenance-bearing semantic working set rather than a flat pile of retrieved prose, with progressive disclosure and explicit context policy.
Our approach
Make the next decision better.
Open standards. Explicit uncertainty. Evidence that survives the handoff.
Entif AI develops open research and infrastructure for intelligence that can be inspected, challenged, and improved. We connect technical questions about meaning and memory to institutional questions about evidence, authority, and accountability.
Evidence with a history
An answer is more useful when you can inspect its sources, distinguish observation from interpretation, and understand what changed. Provenance provides that history; it does not substitute for evaluating the evidence itself.
Open foundations, accountable systems
Rosetta is developed through public specifications, code, and conformance work. Interoperability gives institutions a shared language without requiring them to surrender control of their systems or endorse every claim they exchange.
Progress that can be tested
We distinguish working foundations from proposals, and proposals from demonstrated results. Research, independent scrutiny, and reversible engineering decisions help turn promising ideas into capabilities that earn confidence.