Canvas · the visual workbench
One memory across images, text, molecules, sequences, and behavior.
It is the same associative memory that finds cross-domain text pairs — now with eyes. Images, sequences, and behavior all run through one engine and one memory, so Canvas works as a single instrument rather than three separate tools. A person and the fleet hold joint attention on one visual: segment a region in a microscopy image and the memory retrieves what matches its structure, each answer carrying a receipt you can audit.
- Image
- Text
- Molecule
- Sequence
- Behavior
- + any
retrieved by structure · across modalities
- IMG Prior specimen with the same boundary morphology
- TXT The finding that named this structure, with its citation
- MOL Candidate compound whose motif matches the region
example canvas session · each hit re-verified against its source after scoring
- 01
Segment & share attention
Circle a region, or let it auto-segment. The human's selection and each agent's pointer are separate, mutually visible layers — you co-reference this exact region together, not a chat beside a picture.
- 02
Retrieve across modalities
One region queries every mounted index at once — images, documents, molecular structures, sequences, behavior — each embedded in its own space and ranked without collapsing what came from where.
- 03
Every answer carries a receipt
Each result arrives with its source artifact id and version, re-verified after scoring. Change a source mid-query and the whole answer fails, loudly. Provenance is enforced, not promised.
Confirm or correct a result and the link that surfaced it is reinforced — the next query in the same session already ranks differently. The workbench learns from the work being done on it.
Who we partner with
Experts with unique data
and high discovery value.
We work with a handful of partners in confidential technical domains. If you hold unique data with high discovery value, we'd like to hear what you're working on.
Confidential details only after a secure route is agreed.