RunlineRunline
Platform & SDK

Harness · Run.sh

The Layer 3 runtime the Runner fleet executes on. Runner is who; run.sh is where.

The harness is the runtime the Runner fleet executes on. If a Runner is who does the work, run.sh is where it runs: the sandboxed environment that gives an agent its tools, its identity, and its limits.

What it provides

The harness is what turns a model into a governed worker. It hands the Runner its scoped identity, brokers every tool call, and enforces the boundary the Grid defines. A Runner cannot reach a core, a vendor, or a secret except through the harness, and every tool hop it makes is wrapped by the Grid: authorized, rate-limited, logged.

Where the work lands

Work does not disappear into a chat log. Whatever a Runner does, the result lands on an Item: a state change, an artifact, a draft for review. The harness is the runtime; the Item is the record. That is what keeps an agent's output governable and auditable rather than ephemeral.

On this page