Scientific data
Open the files research actually produces
CZI, ND2, OME-TIFF, NIfTI, DICOM, HDF5, z-stacks, large tiled images, video, and more enter one resource model.
2026.07 research release
BisQue Ultra
A scientific AI workbench for data, models, and evidence.
Open complex scientific files, run tool-guided analyses, inspect every result, and move models from retraining to canary and rollback while preserving the scientific context.
90+ scientific formats Durable runs Gated training Replaceable models
Inspectable work
Keep the answer attached to its evidence
Prompts, tool steps, figures, tables, math, reports, source files, and run events remain part of the same record.
Model lifecycle
Train, benchmark, canary, promote, or roll back
Gold-gated continual finetuning turns model updates into explicit decisions instead of silent weight swaps.
Deployment control
Own the data path and choose the model endpoint
Run locally with Docker Compose, connect BisQue, and point the worker contract at an OpenAI-compatible server.
One scientific record
From difficult file to defensible result, with scientific context intact.
BisQue Ultra unifies the core elements of scientific AI: source data, multidimensional viewing, tool execution, model reasoning, generated artifacts, evaluation, and the operational history needed to return to the work later.
Bring the scientific object
Upload directly or connect a BisQue resource with metadata, dimensions, annotations, and collection context intact.
Inspect before inference
Use tiles, slices, z-scrub, histograms, scalar-volume controls, HDF5 structure, and video previews to understand the data.
Run tools and models
Compose analysis in natural language while specialized tools, code, vision services, and workers execute against the selected resources.
Review and reuse the evidence
Return to the report, figures, tables, metrics, run events, and downloadable artifacts without reconstructing the experiment from memory.
Current product
The workbench now spans data operations and the model lifecycle.
These current interfaces come from the same frontend used by the local stack, with each surface backed by the Go control plane and its durable contracts.
Resources
Scientific data remains a research object across the workflow.
Search, filter, preview, organize, share, rename, restore, and move resources into analysis. The same system handles upload progress and recovery, collections, BisQue imports, direct viewers, and derived pyramids for large images.
- 90+ scientific image formats through the libbioimage service
- Tiled pyramids, z-scrub, scalar slices, physical units, HDF5, and video posters
- Durable upload sessions, resource sharing, folders, and dataset snapshots
Gold-gated training
Model improvement becomes a reviewable, evidence-gated decision.
Freeze a gold set, launch retraining, follow live progress, compare a candidate against the active model, and promote only when the declared gates pass. Canary routing and rollback keep the serving decision explicit after evaluation.
- Durable training jobs, input snapshots, evaluation reports, and model versions
- Per-class floors, regression tolerances, held-out checks, and fail-closed promotion
- Candidate, canary, active, retired, rejected, and rollback states
Research breadth
A shared workbench, with methods shaped by each measurement.
Ultra does not pretend every domain is the same prompt. It supplies a common control layer while domain tools retain their own data contracts, validation logic, and evidence boundaries.
Imaging
Lens keeps multidimensional data inspectable.
Microscopy, neuroimaging, remote sensing, large mosaics, scalar volumes, and time series keep their dimensions and viewing semantics.
Analysis
Language routes into tools, code, papers, and vision services.
Structured results can include rendered math, GFM tables, charts, plots, images, masks, metrics, and downloadable reports.
Materials research
Bounded tools return evidence with explicit limits.
CALPHAD, structure and thermodynamics, kinetics, crystal plasticity, degradation, characterization, and sensor workflows are exposed as evidence-gated research capabilities.
Read the materials research noteBright 4B workload
A 4B-parameter vision model becomes more useful when its evidence remains reviewable.
Bright 4B learns on the unit hypersphere to segment subcellular structures directly from 3D brightfield volumes. Native Sparse Attention, residual HyperConnections, soft Mixture-of-Experts, and anisotropic patch embedding address long-range context, representation stability, adaptive capacity, and confocal geometry. Ultra gives that workload a durable path from volume selection to figures, quantitative tables, and review.
Read the Bright 4B paper
Stage the volume
Open a scientific image or BisQue dataset with dimensional context and prior metadata intact.
Run the method
Execute model-guided work through a recoverable worker contract that streams progress and preserves artifacts.
Interrogate the result
Review masks, figures, tables, prose, metrics, and provenance together instead of accepting a detached answer.
Open architecture
Change the model without changing the scientific record.
The product separates data, run control, worker execution, model serving, and interface state. That makes the system operable, debuggable, and adaptable to a lab's infrastructure.
BisQue and resources
Images, datasets, tables, metadata, annotations, modules, uploads, and collections remain the scientific substrate.
Go control plane
Auth, threads, runs, leases, training state, events, artifacts, and the OpenAPI contract sit on durable Postgres state.
NATS and Deep Agents
JetStream dispatches long work to scalable workers that execute tools, code, paper workflows, and domain services.
React workbench
The interface keeps resources, conversations, streamed steps, viewers, artifacts, training, and administration visible.
Evaluation questions
What teams ask before they bring Ultra into a lab.
Can BisQue Ultra run locally?
Yes. Docker Compose launches Postgres, NATS JetStream, the scientific image service and conversion worker, the Go control plane, a Deep Agents worker, and the React frontend. You provide a compatible model endpoint.
Does BisQue Ultra require an existing BisQue deployment?
No. The local stack supports direct uploads and local resources. An existing BisQue deployment is optional and adds shared images, datasets, metadata, annotations, modules, and institutional workflows.
Which scientific files can it open?
The image service decodes more than 90 formats, including CZI, ND2, OME-TIFF, NIfTI, DICOM, multi-page TIFF stacks, HDF5 workflows, large tiled images, and video.
Can a lab use its own model endpoint?
Yes. The worker uses an OpenAI-compatible contract and can point at Ollama, vLLM, or another compatible server. Specialized vision weights and scientific checkpoints remain operator-provisioned.
Is the materials capability production-ready?
The current materials surface is an evidence-gated research capability. Production promotion still requires the designated live traces, ledger qualification, production sandbox and isolation evidence, external evaluation thresholds, and an attested promotion envelope.
Release notes
Read the product through the systems that make it credible.
The launch brief gives the complete release view. The engineering notes go deeper on imaging, model operations, materials evidence, the BisQue substrate, and interface design.
Launch brief
BisQue Ultra 2026.07
The 2026.07 research release brings scientific file and viewer infrastructure, durable analysis, gold-gated training, and evidence-aware domain tools into one workbench.
July 13, 2026 ยท 2026.07The scientific image engine behind BisQue Ultra
A technical tour of the image service, conversion worker, resource model, and multidimensional viewers that let Ultra work with the files scientific instruments actually produce.
From retraining to rollback: the GoldGate model lifecycle
Why a scientific workbench needs more than a training button, and how GoldGate makes model versions, evaluation failures, canaries, and rollback visible.
Evidence-aware materials research in BisQue Ultra
A research note on typed materials tools, deterministic validation, provenance, and the release boundaries that keep scientific capability separate from unsupported claims.
BisQue Platform: storage, visualization, analysis, and extensibility
A guided explanation of what the BisQue platform already knows how to do, and why storage, visualization, metadata, modules, and deployment give BisQue Ultra real scientific weight.
Why the BisQue Ultra frontend looks the way it does
A design-language article on the BisQue Ultra frontend. It explains the reasoning behind color, typography, spacing, interaction design, streaming behavior, and evidence handling, using the product's actual values and constraints rather than generic design slogans.
Work with us
Bring us the scientific workflow that deserves a connected system.
We are opening BisQue Ultra to research teams working with complex scientific data, domain models, and reproducibility requirements. Tell us what you measure, what you need to run, and where the evidence currently breaks apart.