Product updates, new features, improvements, and fixes as we ship them. Subscribe via RSS to stay updated.
Chat now holds steady as usage grows. Sessions persist across instances through a shared real-time layer, so conversations no longer drop when traffic spreads across servers. Scheduled autoscaling matches capacity to your busiest hours and scales back when things are quiet.
Administrators can now ship a chat workspace that stays locked to its approved configuration. Settings are fixed at startup and cannot be changed from the interface, so security choices hold on every restart. Higher-risk capabilities like code execution, open sharing, and custom voice prompts are turned off by default.
Teams now get a managed chat workspace that carries Swept branding and connects straight into the rest of the platform. It runs on a maintained build of Open WebUI, so the chat experience stays familiar while Swept's controls, navigation, and model access sit on top. There is no separate setup or fork to keep up with.
Your applications can now reach top models through one consistent, OpenAI-compatible endpoint instead of wiring up each provider on its own. The gateway routes to Google Gemini and Anthropic Claude, runs entirely on private networking, and issues scoped keys so every app gets its own access and budget.
Chat can now pull in current information from the web and work through multi-step questions before it answers. Search is available on both Google Gemini and Anthropic Claude, with extended reasoning on Gemini. Answers stay grounded in what the model finds at the moment, not memory alone.
Compliance teams can now produce a complete, examination-ready package whenever a regulator asks. Each package gathers the relevant risk records, approvals, and evidence into a single export, with a full audit trail behind it. Output comes as a combined PDF, scoped to the right committee.
You can now set token budgets for individual users and groups, then watch spend against them in real time. In-app alerts flag usage as it climbs, and a simple forecast shows where the month is heading. It all lives in the workspace, so there is no separate spreadsheet to keep.
Supervision events can now flow straight into your own tools as they happen. Bring events in from external sources, and push them out through signed callbacks your systems can verify. This makes it simple to route live AI activity into the monitoring and alerting you already run.
Teams can now sign in with their own identity provider through OpenID Connect, including Microsoft Entra, across Workbench, Cloud Lock, and chat. Access is granted by group and by user, so people see only the apps and features meant for them. Credentials stay with your provider.
We keep the platform current against newly disclosed vulnerabilities as part of regular maintenance. Recent work patched dozens of issues across the underlying frameworks and libraries. Updates ship continuously, so deployments stay on supported, secure versions.