Checksum has launched the API Agent, a continuous testing agent that generates and maintains journey-based tests for backend APIs.
The agent builds multi-step tests that mirror how a product actually uses its API, keeps them current as the API changes, and runs them in a team’s existing pipeline. It closes the gap that opens when AI coding tools add endpoints faster than engineers can write tests to cover them.
Spec-based generators stop at the endpoint. They confirm a route exists or matches a schema, and miss the behavior that actually breaks in production: an ID passed between steps, a state change that has to hold across a sequence.
The API Agent generates stateful, multi-step journeys instead, capturing dynamic values like IDs and tokens at runtime, covering error cases alongside successful flows, and producing the tests as standard PyTest that engineers can read and own. Across Checksum’s platform, teams see 82% lower test failure rates than they do maintaining suites by hand.
“Every team knows it should have API tests. The reason it doesn’t is that the tests rot faster than they catch anything, so the suite becomes one more thing to maintain instead of something the team trusts,” said Gal Vered, CEO of Checksum. “We built the agent to carry that maintenance, so coverage grows with the API instead of decaying against it.”
Engineers can now generate a complete API journey (create a resource, capture the returned ID, read it, update it, verify the result) as readable PyTest, with the values wired between steps automatically. Previously that meant writing and repairing every flow by hand.
The agent starts from a spec or captured traffic, with no repo connection required, and reaches meaningful coverage across a large API surface in days. When an endpoint or schema changes, it identifies the affected journeys and proposes updates through a reviewable session, so engineers approve changes instead of hunting down broken tests. For teams already using Checksum’s end-to-end agent, API and end-to-end coverage now live in one platform.
