Supravision
Human oversight from any device, in real time
See what your agents are doing right now. Approve or deny suspended actions from your phone. Panic-stop a rogue agent on a remote server. Review held output before it reaches the agent. Supravision gives you a live control plane over every agent session, from anywhere.
Your Agents Run Where You Aren't
Agents run on servers, in CI pipelines, on employee workstations, on your kid's laptop. You can't be logged into every machine watching every session.
But when something goes wrong — an agent making unexpected API calls at 2am, a child's homework assistant retrieving inappropriate content, a financial agent about to authorize a suspicious transaction — you need to see it and act on it from wherever you are.
One Dashboard. Every Agent. Every Device.
The Supravision dashboard gives you a unified view of all active agent sessions across all registered devices. Real-time session status, live attestation streams, budget consumption, and pending approvals — all synchronized within seconds.
Register devices once. Monitor everything from anywhere.
Cross-Device Control
Register multiple devices. Send commands from any of them. Pause a session on your server from your phone. Resume from your laptop. Every command is time-bounded — stale commands expire and never execute.
Remote Approval Workflows
When enforcement suspends an action for human review, you get a notification on your connected devices. Preview action details. Tap approve or deny. Set reuse rules for identical future actions.
Output-Hold Review
When content needs review before reaching the agent, the output is held and you’re notified. Preview. Deliver or suppress. The agent never sees it until you say so.
Emergency Panic Stop
One tap. Agent stops immediately. Child processes terminated. Network connections severed. Attestation chain sealed. Forensic record preserved. Available from any connected device.
Guardian Mode
Configure a guardian profile for child safety. Screen time budgets. Content filter categories. OUTPUT_HOLD triggers on filtered content. Guardian reviews from their device. Session auto-stops at budget expiry.
Managerial & Compliance Mode
Same output-hold mechanism, different triggers. Managers require review for financial actions above a threshold. Compliance officers hold data exports for regulatory review. Configure trigger criteria in the reviewer profile.
Configurable Alerts
Rules that fire on specific patterns: 3+ blocked actions in five minutes, spend exceeding 80% of budget, content filter triggers, SUSPEND frequency spikes. Dashboard, email, or webhook delivery.
Remote Task Injection
Not just monitoring — commanding. Send a task to an active agent session through the dashboard. The injected task goes through the full enforcement pipeline before the agent receives it.
Developer on the Go
Your research agent is running on a cloud server. From your phone at lunch, you see it’s burning through budget. You reduce the spend limit. Later, it hits a suspicious pattern — the engine suspends and notifies you. You review, deny, and set a new constraint. All from your phone.
Parent at Work
Your child is using an AI homework assistant. The assistant retrieves an article flagged by the content filter. You get a notification. You preview on your phone, determine it’s appropriate, and tap approve. At the 45-minute mark, the session auto-stops.
Compliance Review
An agent handling financial data triggers an output-hold on a data export. The compliance officer reviews, approves delivery. The entire review — who, when, what they decided — is recorded in a signed attestation.
Related