Changelog
What changed in CertiQA
Release notes for the platform, newest first. Each entry lists what shipped and what it lets you do.
Cost per successful task, A2A conformance and local step debugging
Compare model and prompt configurations by cost and pass rate, test agent-to-agent protocols, and debug a broken test one step at a time on your own machine.
Pareto frontier and cost gate
Plot configurations by cost per successful task and pass rate, see which are dominated, and gate a pull request in CI when a change lowers the pass rate, raises cost past a ceiling, or is beaten by another configuration.
Agent-to-agent (A2A) conformance
Conformance testing for multi-agent setups with A2AAdapter and A2AConformanceValidator. Measure how context compaction affects multi-session journeys, and run long-term memory poisoning attacks (OWASP LLM01 and LLM06).
Local step debugger
Debug a failing test in a headed browser on your own machine with the CertiQA CLI: step over actions, retry a step with changed parameters, and try selectors against the live page. Generate a 15-minute ticket from the case, copy the command, and run it in a terminal. Nothing from a debug session is recorded as a run, and debugging is not hosted.
Agent Studio and release reports
First-class AI agent testing, with tool calls mocked in flight and a signed report for every CI run.
Tool-call mocking
Intercept REST and Model Context Protocol (MCP) tool calls during CI runs, so tests do not touch live databases or external APIs.
Release reports
Every CI test run produces a SHA-256 stamped report with Wilson 95% confidence intervals and calibrated LLM-as-a-judge rubrics, so a change can be checked for tampering after the fact.
Try the latest release on your own agent.
Run your first agent test in minutes.