Glossary

Domain terms used throughout the documentation, alphabetical.

Annotation. A bounding-box sequence with keyframes and a label, stored as one row in the annotations table. Either type or object; see Concepts > Annotation model.

BullMQ. The Redis-backed job queue library. Backs summarization, claim extraction, claim synthesis, and detection.

Claim. A hierarchical assertion extracted from a video summary. Carries text, gloss, modality flags, and optional parent. See Concepts > Claims model.

Claim relation. A typed edge between two claims. Type comes from the persona's relationTypes.

Cross-user import. Import where the JSONL was exported by a different user, detected via metadata.exporterUserId. Triggers id regeneration and gloss remapping.

Detection. Open-vocabulary object detection driven by the persona's entityTypes. See Guide > Detection.

Fovea. Flexible Ontology Visual Event Analyzer. The project.

Fusion. Combining audio transcription and visual summarization into a single summary. Strategies are sequential, timestamp_aligned, native_multimodal, hybrid.

Gloss item. One entry in a claim's gloss array. Type is one of text, objectRef, typeRef, annotationRef, claimRef.

Information need. Free-text statement on a persona of what the persona is trying to learn from a video.

Keyframe. A frame at which the user has explicitly placed a bounding box. Frames between keyframes are interpolated.

linkType. Column on Annotation recording whether an object annotation is linked to an entity, event, time, or location. See Guide > Annotations.

Model service. The FastAPI process hosting VLM, LLM, detector, tracker, and audio adapters. See Concepts > Model service.

Multi-user mode. FOVEA_MODE=multi-user. Each user has private personas, ontologies, world state, annotations, summaries, and claims. The default for production.

Object annotation. An annotation whose label is a world-object id (entity, event, time, or location). Discriminated by linkType.

Ontology. The four type lists belonging to a persona: entityTypes, eventTypes, roleTypes, relationTypes.

Persona. Named role plus information need. Owns one ontology. Scopes annotations, summaries, and claims.

Single-user mode. FOVEA_MODE=single-user. The single seeded user owns everything. Used for local development and demos.

Synthesis. Re-running claim extraction against a revised summary. Distinct from initial extraction.

Type annotation. An annotation whose label is an ontology typeId. Requires personaId.

VLM. Vision-language model. Used for video summarization.

Wikibase / Wikidata. External linked-data backends. Online mode hits public Wikidata; offline mode hits a local Wikibase.

World state. Per-user collections of named entities, events, times, locations, and relations. Persona-independent.