Links the same person or vehicle across cameras and answers where, when, and who-was-there questions with a provenance trail, refusing to link when it is not confident.
Closed source, inquireFleetmind ingests any tracker's JSONL event stream and links identities across cameras with an asymmetric, precision-over-recall transit-time-prior linker, favoring a missed link over a wrong one.
Linked history persists to a bounded SQLite store that the project has shown survives restarts and stays byte-bounded over a simulated month, and answers provenance-carrying queries: where a person or vehicle is, their history, who was at a place, journeys, and co-occurrences.
It ships a real integration with Frigate, the open-source NVR project: a live MQTT bridge maps Frigate's own recognized license-plate or face outputs directly into fleetmind's identity field, upgrading a transit-prior guess to a certain link for free.
Linking precision 1.0, recall 0.377 (69 true same-person track pairs, 26 linked, 0 wrong); the whole site's memory footprint stayed at 60.0 KB; cam0-to-cam1 journeys were recovered in the 37.85 to 53.65 second range.
Precision 1.0 with zero wrong links in every tested condition, across both a ground-truth-track leg and a real YOLO11n plus ByteTrack pipeline leg, at up to 40 percent identity coverage. Recall in this evaluation (0.35 to 0.41) comes entirely from the exact-identity path; the transit-prior mechanism, the project's actual novel contribution, fired zero links here, because WILDTRACK's cameras overlap rather than forming a non-overlapping transit network, and the project's own results file says so directly rather than let the recall numbers imply otherwise.
An outsider can download the published benchmark dataset and try the demo Space to see the linking and provenance behavior described above. The WILDTRACK real-data evaluation and the Frigate integration are reported from the project's own files; the source code is not public, so they are not independently re-runnable from outside yet.