Harness Designer

The Harness Designer lets you visually edit your robot's agent harness configuration.

Opening the designer

Robot detail → Harness tab → Edit (pencil icon)

Pipeline view

The designer shows a visual flow of the harness pipeline:

[Input] → [Pattern] → [LLM] → [Cost Gate] → [Consent] → [Action]

Each block is tappable. Drag to reorder layers.

Design panels

Three panels at the bottom of the pipeline:

Pattern panel

Choose the execution pattern: - Single Agent Supervisor — default, one agent + supervisor - Initializer / Executor — two-phase for complex tasks - Multi-Agent — parallel branches (server-class only) - Reactive — event-driven, low latency

Memory panel

Choose the memory backend: - Working — in-memory, fastest - Filesystem — persists to disk across sessions - Firestore — fleet-visible, cloud-synced

Configure overflow strategy: Summarize / Drop oldest / Error.

Security panel

Optional OPA guardrail and telemetry export.

Applying champion configs

When a champion config is pending, an amber banner appears at the top of the Harness screen:

🏆 New champion available — lower_cost · 0.6541

Tap Apply to merge the champion tunables into your current config. Tap Review to see a diff before applying. Tap Dismiss to defer.