Runtime — Dry Run

One flag.
Nothing executes.

Pass dry_run: true on any runtime call. Every write your agent tries to make is intercepted before it reaches the provider. Reads pass through normally. Your agent sees exactly what would happen — real responses, real data, zero production impact.

Not a simulation.
The same execution path.
Writes intercepted, not skipped.

API Docs → Get API Access

How it works

Same execution path.
Writes intercepted at the fence.

Dry Run is not a separate test environment. It runs the identical execution path as a live call — governance fires, credentials resolve, the idempotency fence checks. The only difference: write operations are caught at the action layer before the provider API is called.

dry run call

# Exact same call as live — one flag added

POST /v1/runtime/actions/run

X-API-Key: sk_live_...

{

"verb": "shopify:cancel_order",

"args": {

"order_id": "gid://shopify/Order/1234"

"dry_run": true

}

# Response — intercepted before provider

{

"success": true,

"dry_run": true,

"intercepted": true,

"would_have_executed": "shopify:cancel_order",

"governance_status": "require_approval",

"execution_id": "exec_dry_7xk2p"

}

Reads pass through normally

GET operations — lookups, data fetches — execute against the real API. Your agent gets real responses, real data. Only writes are intercepted.

Governance still fires

Guardrails check the verb against your policy even in dry run. You see exactly what the governance decision would have been — approved, blocked, or escalated — without any execution.

Audit trail still writes

Every dry run call is logged with dry_run: true in the audit record. You have a full trace of what your agent tried to do — even before it went live.

Idempotency fence still checks

Pass an idempotency_key in dry run and duplicate calls are still caught. Same behavior as live — so you can test retry logic safely.

When to use it

Before your agent
touches production.

Dry Run is built for every stage where you need to validate agent behavior without risk. New agent, new integration, new policy — run it dry first.

New agent deployment

Run your entire agent workflow in dry run before flipping to live. See every action it would take, every governance decision, every credential it would need. Ship with confidence.

Policy validation

Changed your guardrails policy? Run the same agent calls in dry run to verify the new policy behaves as expected — before it blocks or approves things in production.

Regression testing

After updating your agent's logic or adding new integrations, run the full workflow dry to catch any actions that would now violate policy or fail credential resolution.

One flag.Nothing executes.

Same execution path.Writes intercepted at the fence.

Before your agenttouches production.

Ship your agent.After you know it works.

One flag.
Nothing executes.

Same execution path.
Writes intercepted at the fence.

Before your agent
touches production.

Ship your agent.
After you know it works.