Charge per agent run on delivered outcomes, not flat subscriptions

Four primitives that fix outcome-based agent billing

Each one closes a margin gap that subscription tools and observability-only stacks can't. Wire them in once and your agent has commercial superpowers from the first run.

Per-agent cost cap in one POST

Out-of-the-box SDK in Node.js and Python. Configure the per-agent cap once; Rev rejects over-cap runs and reports the cost side-by-side with the outcome bonus.

// Configure a per-agent cost cap on the agent billing tier
const tier = await rev.pricing.createTier({
  agent_id: 'research-agent-v2',
  base_fee: 0.10,          // $0.10 per agent run
  token_rate: 0.00003,     // per token consumed
  outcome_bonus: 4.00,      // $4.00 per task the user acks
  cost_cap_usd: 0.50,       // per-agent run ceiling
  currency: 'USD',
});

// Meter every agent run event — returns the OK billable event
const event = await rev.meter.emit({
  agent_id: 'research-agent-v2',
  customer_id: customer.id,
  tokens: inputTokens + outputTokens,
  outcome: 'pending',
});

// Resolve the agent outcome — bonus posts on success, refunds on failure
if (userAcked) {
  await rev.outcomes.resolve(event.id, 'success');
} if (taskFailed) {
  await rev.outcomes.resolve(event.id, 'failure');
}

Rev vs observability-onlyagent stacks

Stripe metered billing, Helicone, and Langfuse observe or charge on schedule. They can't enforce per-task cost caps or reconcile a hindsight outcome. This is the gap Rev fills for AI agents.

The common pattern: keep Stripe for charging, Helicone or Langfuse for trace observability, wire Rev for outcome-based metering. The services compose — Rev sits in front of charging as the pricing engine, and behind your agent runs as the per-task cost-cap layer.