Daily verification
Enroll a model artifact or a package and follow its scheduled checks. See availability, file-size and SHA-256 results with dated evidence. An unavailable check stays visible as an uncertainty.
MODEL X / PRODUCT
Visibility for the models you depend on.
Monitor artifacts, inspect security risks, and build a verifiable path to recovery.
Create account Explore product ↓THE PRODUCT
From a model URL to a clearer picture.
Changes with context
Where the source exposes the required information, follow repository revisions, model-card and license changes, configuration changes, and dependency notices. Reports provide before-and-after details when available.
A history you can inspect
Open a model’s monitoring history to compare dated reports and investigate changes. A successful check describes what was verified at that time; it does not promise that a remote host will remain available.
Private JPEG snapshots
Preserve the appearance of a page you have permission to capture. Request a snapshot or enable daily capture; changed images are added to its private history. Snapshot requests are queued and may take several hours to complete. Daily snapshots are scheduled approximately every 24 hours and may also be delayed.
Optional Pro recovery
Authorize preservation of an eligible file or package. Stored copies, recovery drills, and download access provide evidence that a retained copy can be recovered. Protection begins after storage and a passing drill.
Pro behavioral audit
Use Kohere-D behavioral audit evidence to examine a model’s responses within the configured audit scope. Read it alongside integrity and monitoring reports. It is not a blanket safety certification.
Daily file inspection
Daily checks inspect configuration, tokenizer, generation, dependency files and custom Python code without executing the model or its code. Coverage depends on the files and metadata exposed by the source.
Enrolled Baseline
Compare the monitored metadata with the enrolled revision. See how the available metadata differs from the revision enrolled for monitoring.
SECURITY GATE
Security findings you can investigate.
PASS Configured checks passedREVIEW Findings or coverage need reviewBLOCK Blocking indicators found
Python & JSON inspection
Flag risky Python execution, remote-code loading and network access. Inspect collected JSON configurations for malformed structure, duplicate keys, custom-code settings and suspicious file references, without executing model code.
Downloaded-file malware scanning
Inspect enrolled artifact bytes with antivirus signatures, supported Pickle/PyTorch content checks and Safetensors structure checks. Each report shows scan coverage.
Dependency vulnerability checks
Check collected, exactly pinned Python packages against known OSV security advisories. See affected dependencies and advisory references.
Baseline & revision checks
Compare available revision evidence with the enrolled baseline and follow changes between scheduled security assessments.
Security change alerts
Highlight newly observed risks and reduced inspection coverage in your dashboard. Repeated findings stay visible without creating another change alert.
Dated inspection history
Review the recorded decision, findings, affected files and available source locations for each inspection. See when configured checks return to PASS.
Security inspection is included across Free, Starter and Pro. Findings appear after scheduled daily checks and are retained in each model’s inspection history. Alerts are shown in the dashboard.
PASS describes the configured checks, not a guarantee that a model is malware-free. BLOCK identifies blocking indicators for your review; Model X does not enforce deployment blocking. File, source and scanner limits can leave checks incomplete.
See supported checks and scanning limits →HOW IT FITS TOGETHER
Observe. Preserve. Verify.
- 01 · EnrollChoose the source and artifact or package. Add an expected SHA-256 when available.
- 02 · ReviewInspect daily results, notices, and dated history in your workspace.
- 03 · RecoverOn eligible Pro accounts, authorize a copy and review its recovery drill before relying on it.
CLEAR BOUNDARIES
Know what your evidence means.
Byte integrity, repository metadata, page appearance, and model behavior are different kinds of evidence. A matching hash does not establish the quality of a model’s answers. A metadata notice does not by itself prove the weights changed.
Recovery depends on permission, source access, and available shared capacity. Pro allows 100 GB per account, with a 4 GB limit per recovery file or ZIP and up to 64 files per package.
YOUR NEXT STEP
Your models. Your workspace.
Start with one model on the Free plan.