Questions & answers

An always-on machine — a server, a workstation, a VM — where an AI agent runs continuously and does real work in your tools: mail, documents, spreadsheets, internal systems. Harness is not the agent itself; it is the caretaker that gives that agent a queue, rules, health checks and a memory of everything it did.

Early access targets agents that can run shell commands and use APIs on a machine you control. If your agent can be scripted, Harness can supervise it. Integration guides ship with onboarding, and we help the first teams connect personally.

It stops and asks. You define guardrail classes — spending money, deleting data, contacting customers, touching production — and any step that crosses one waits for an explicit human approval. The request, the decision and the outcome all land in the audit log.

Your agent keeps running on your machine, in your accounts. Harness stores the coordination layer: tasks, schedules, approvals and logs. We never train on your data and you can export or erase your history at any time.

Early access is free while we onboard the first teams. Paid plans arrive with general availability, and early-access teams get a founding discount and generous notice before anything starts costing money.