HyperView documentation
HyperView is an agent-native workbench for inspecting datasets, comparing representations, and sharing evidence. People and coding agents operate the same runtime-owned state through the UI, CLI, and API.
Core concepts
HyperView keeps the product surface, command line, and agent interface aligned around the same workspace model.
Datasets, embeddings, layouts, selections, panels, and jobs live in the workspace—not in a browser-only copy.
Inspect samples, maps, metrics, and purpose-built evidence together in one synchronized workspace.
The CLI and API expose the same state that people see, so agents can discover, compose, and verify work.
Package prepared evidence as a read-only Space, or connect a live runtime when new computation is required.
HyperView Spaces
Each Space is a prepared, shareable HyperView artifact—not a screenshot or stripped-down mini app. Open one to inspect the complete read-only workspace.
Read-onlyProduct retrieval workbench
Does the model find the right product—not just a plausible category match?
Compare prepared text and image retrieval cases across two model spaces, with linked results and catalog topology.
Read-onlyRegion-level retrieval audit
Does the exact region reach the operator's first screen?
Inspect the source scene, boxed ground truth, and aligned Top 5 results over the same candidate pool.
Read-onlyTyped product search audit
Does the exact SKU reach the shopper's first screen?
Audit whether decisive product attributes survive multimodal retrieval, then inspect the result in catalog context.
Read-onlyCreative asset retrieval
Which existing logo best satisfies a detailed creative brief?
Compare prepared brief results, exact-asset ranks, and archive coverage without relying on manual folders or tags.
Read-onlyRemote-sensing archive QA
Do retrieved neighbours preserve land-use identity and avoid costly confusion?
Inspect exact and parent-class consistency while comparing how two model maps organize the same aerial archive.
Read-onlyReview-queue operating-point audit
Is one extra catch worth five false reviews and six more queue slots?
Compare a fixed review rule across two models and inspect the concrete gains, misses, and operating costs.
Live runtime
Static Spaces preserve prepared evidence. Connect a runtime to create datasets, run providers, recompute layouts, and keep agent actions in workspace state.