Agents and workflows
Agents, sessions and recovery
A task history is useful only when it can be tied to the permissions and effects of the work performed. SAVRN Cloud uses Work Orders to illustrate this distinction, then identifies the checkpoints, cancellation and reconciliation needed when a real agent encounters an uncertain outcome.
SAVRN Cloud · Coming soon · Public preview · Browser-local simulation · No live compute or payments
Use a Work Order as the example
Open Work Orders and inspect a small task. Follow its decision in Approvals, then run the supported local workflow. Review the resulting sample artifact and usage entry. This demonstrates a bounded task lifecycle; no autonomous agent calls external tools or performs real research.
The application keeps state in this browser. Reloading can restore local records, but it does not resume a server worker or establish durable execution across machines. Treat an apparent session history as demonstration data.
Ask what actually happened
A useful agent record identifies the task, permitted sources, tools, budget, human decisions and final artifact. A status message is weaker evidence than the retained effect. Distinguish a proposed action from an approved action, and an approved action from a completed effect.
Use Accept deliverable or Return for revision where the Work Order exposes those review actions. Review measures whether the deliverable satisfies the request; it does not retroactively prove external execution.
Before connected services launch
Implement scoped tool permissions, bounded retries, checkpoints, cancellation and safe recovery after uncertain outcomes. Preserve idempotency or reconciliation where effects can duplicate. Approval gates must bind to the action being executed. Secrets and private sources require narrow access. A failed task must retain enough redacted evidence for diagnosis without exposing unrelated customer data.
Build toward the work that matters.
Tell SAVRN what your institution needs to run, who reviews the results and where its data must stay. That workload defines the next service to qualify.
Discuss an AI project