01What is AiOverload, in one sentence?
AiOverload runs AI agents as deterministic, event-sourced workflows on your own infrastructure, with a risk-scored policy gate, recorded human approvals, default-deny gateway tools, and a replayable event log.
02Is it another agent framework?
No. Use the agent framework and harness that fit your work. AiOverload is the governed-execution layer around them: manifests, orchestration, admission decisions, tool grants, operator input, and evidence.
03How is it different from a workflow engine like Temporal?
A workflow engine provides orchestration primitives; AiOverload focuses on the agent-specific governance layer: named agent and tool manifests, a reflex risk gate, held steps with human input, gateway-routed tool decisions, and replayable run state.
04What happens when a step looks risky?
The step parks as step.held. The reflex gate escalates above deterministic thresholds — 0.70 for steps and 0.45 for gateway-routed tool calls at the stamped commit. An operator can approve with input or reject the step; the reason and decision are written to the event log.
05Where does run state live?
In ordered events on NATS JetStream. The event log is the runtime, not a report generated afterward. Git and the forge hold desired registry and task state; aioctl replay re-verifies a run against its definition and events.
06Does default-deny cover every tool an agent can call?
No. It covers the declared MCP tool plane routed through agentgateway. Harness-native tools such as built-in Bash, Edit, and Write can bypass that path today. The security page names this as an open gap rather than implying complete coverage.
07Is it open source? Can I self-host it?
Yes and yes. The engine is Apache 2.0 and is designed to run on your infrastructure. You can use, modify, self-host, and redistribute it under the license conditions. The engine repository and its documentation are the authoritative release artifacts.
08Is there a hosted service or pricing?
No hosted offer or pricing is claimed on this site. The project is self-hosted and still preparing its public engine release. Any future commercial artifact should be described only when it actually exists.