v3.7.0

Released September 15, 2026

This release bundles a set of reliability, billing-accuracy, and data-integrity fixes across models, token allocations, and content protections.

Forward‑only upgrade. This release includes a database change and cannot be rolled back to the previous version once applied. Self‑hosted and Ask Sage Edge deployments should plan accordingly; managed (SaaS) tenants are upgraded by Ask Sage.

Confirming this release. Oversized requests are rejected up front with a clear message instead of failing silently.

🔧 Improvements & fixes

More reliable government model access. Resolved an issue that could cause certain government models to return errors, so requests reach the intended model dependably.
Consistent model availability across app versions. Model catalogs and allowed-model settings now stay in sync between the classic and v2 experiences, so the same models are available in both.
Fairer handling of oversized requests. Requests that exceed the allowed size are now rejected up front and are never charged, so you’re not billed for a response that couldn’t be produced.
Accurate monthly token resets. Token allocations now reset exactly once per month and never leave a balance stranded below your usage.
Stronger protection for controlled content. Access checks for sensitive and controlled content are now enforced at the point of access, with clearer messaging when access is blocked.
Names display correctly. Stopped altering names when they’re saved and corrected how they’re displayed, so names containing characters like apostrophes and ampersands render properly.
Clearer admin-log messaging. Improved error handling in the activity-log views, including a clear blocked-state notice when access isn’t permitted.

🔒 Security & platform

Ask Sage services are delivered as FIPS‑hardened, digest‑pinned, and cosign‑signed images within the FedRAMP Class D authorization boundary, and are continuously rebuilt against the latest patched base images as part of Ask Sage’s continuous‑monitoring commitment.