Personalization has hit a ceiling.
Segments, demographics and last product purchased can tell you who a customer resembles. They cannot tell you what a customer is trying to do, and relevance lives entirely in that gap. Intent, deep behaviour and live context are what separate an experience a customer acts on from one they dismiss, and almost none of it exists as a field in a CRM.
The engagement maturity curve
- Geography, age, gender
- Centralized customer profiles
- Product holdings and history
- Product adoption and channel preference
- Dynamic attributes
- Trigger-based interactions
- Stated intent and needs
- Persistent per-person memory
- Goals, preferences, life circumstances
- Reasoned, explainable recommendations
- Unified AI and data alliances
A living model of an entity, not a static record.
A Chronicle is a persistent, evolving knowledge graph of a single individual or entity that continuously updates with goals, preferences, behaviours and life circumstances.
The deeper the story, the better the decision.
A Chronicle tells the story of an individual or an entity, and that deeper view is what produces better decisions and better outcomes.
Chronicles AI for risk and fraud
Risk teams work from snapshots: a score computed last month, an alert raised after the money moved. A Chronicle holds the account’s whole trajectory, so the same signals read as a direction rather than a reading — and the pattern shows up while it is still forming.
How tech is breaking the rules of biology
Our founder Hossein Rahnama sits down with Emily Chang to talk through the idea underneath Chronicles: that a person’s accumulated context — their goals, their history, the way they actually make decisions — can be modelled, and what it means when it can.
The Chronicle pipeline: construction, learning, reasoning.
Every Chronicle runs the same three stages. Raw signals are built into structured situations, those situations accumulate into a model of the individual or entity, and questions are answered against that model, grounded in the specific situations that support the answer.
Construction
Learning
Reasoning
Situations, typed on a formal ontology
A situation is a node with formal relations to the entities that constitute it. Using an upper ontology — DOLCE Ultralite — rather than an ad-hoc schema is what lets situations captured from different channels be compared, merged and reasoned over together.
One graph per time segment, concentrated over time
Construction initialises a Chronicle from a profile and the first segment of channel data. Each channel — image, text, audio — passes through its own content-embedding pipeline, and a graph-embedding pipeline places the resulting situation graph in the same space so segments can be compared. As segments repeat, situation attention and identity inference concentrate the graph: reinforcing what recurs, decaying what does not, and revising the identity model as the entity changes.
Chronicle learning across one time segment. The loop is the point — each new segment re-enters the same graph rather than replacing it.
Answers that carry their evidence
A question retrieves relevant concepts by semantic similarity, traverses the graph for the situations related to them, and a fine-tuned model composes the answer from what was retrieved. Because every node holds a description and a link back to its source artefact, the answer arrives with the situations that produced it attached — which is what makes it reviewable rather than merely fluent.
Chronicle reasoning. The three retrieval steps all read from the reason-ready chronicle; the model composes, it does not invent.