Overview
AI Agents now show you what they cost to run, before and after execution. You get an AI Credits estimate when testing a behavior, and the real consumption broken down by node once the agent runs, giving you full visibility into the cost of every configuration.
What's new?
- Test Behavior cost estimate: see how many AI Credits a tested configuration would use before publishing it. Testing does not consume AI Credits.
- Execution Summary: see the total AI Credits consumed by each agentic run, right in the Summary.
- Cost per node: understand how AI Credits were distributed across individual execution nodes.
How does this help?
- Safer configuration decisions: test an agent's behavior and see the estimated cost before pushing it to production, without spending any AI Credits.
- Predictable consumption: track the real cost of each execution in the Summary, no manual math required.
- Spotting expensive nodes: see exactly which node is consuming the most AI Credits and adjust the agent's configuration accordingly.
Use cases
Builder testing an IT ticket triage Agent
Before publishing an agent that classifies and prioritizes tickets opened on the Service Desk, the Builder runs Test Behavior with different prompts and connected tools, such as web search and IDP for reading attachments. The AI Credits estimate shows up for each tested variation, without spending credits. The Builder compares the options and picks the configuration that classifies tickets correctly at the lowest cost before rolling the agent out to the support team.
Pipe Admin monitoring an invoice analysis Agent
An agent reads invoice attachments, extracts amount, vendor, and expense category, and updates the card fields automatically. After a few weeks in production, the Pipe Admin opens the Summary for recent runs and spots that the document reading node, IDP, is consuming most of the AI Credits, far more than the category classification node. With that data, they adjust the flow to trigger IDP only when the attachment actually needs complex table reading, cutting consumption without losing accuracy.
Operations Leader reviewing the cost of an expense approval Agent
The agent evaluates reimbursement requests, checks the requester's history, and decides whether to auto approve or escalate to a manager. The operations leader uses the Total Cost in the Summary to compare average cost per run over the month and check whether the volume of manual escalations lines up with AI Credits consumption, adjusting the auto approval criteria when cost per run climbs without any gain in accuracy.
Builder tuning a customer response Agent with Websearch
An agent answers customer questions by checking order history and, when needed, looking up current public information, such as delivery status from external carriers. The Builder uses cost per node in the Summary to confirm the Websearch node only fires when it actually needs external data, not on every question, avoiding unnecessary AI Credits consumption.
Most relevant for
- AI Agent Builders
- Pipe Admins
- Operations Leaders
- Technical Product Managers
Plans and availability
This feature is rolling out gradually. Testing a behavior in Test Behavior does not consume AI Credits, only real production runs are counted.
