The trial includes 4,000 credits and four concurrent runs, so a team can generate functional and adversarial suites from its own agent's docs and endpoint, with code optional, and gate CI on evidence-backed verdicts without booking a demo.

SAN FRANCISCO & NOIDA, India: #AIAgents--TestMu AI (formerly LambdaTest), the world's first Agentic AI-powered Quality Engineering platform, today announced the general availability of Agent Assurance. Agent Assurance tests AI agents that take actions, such as calling tools, writing files, and changing records, by checking what a run actually changed rather than what the agent says it did. Any team can now create a TestMu AI account and run it against its own agent without booking a demo first.

Agent Assurance launched on August 17, 2026, behind a waitlist and a guided onboarding session. Since then, Rook, the CLI that runs Agent Assurance, has shipped a steady run of releases, shaped by feedback from early teams.

Agent evaluation has mostly graded the final reply, and that reply can be wrong about the agent's own work. In CXBench, a benchmark of LLM agents in regulated customer-experience workflows published in October 2026, 80.1% of runs that reached the right outcome still failed the process checks, and only 13.8% passed both. An agent that calls the wrong tool or skips a required step can still report success. That is why Agent Assurance checks the effect of a run, covering the tool calls the agent made and the records, messages and artifacts it changed, as reported by hooks the team connects, before it issues a verdict. Anything it cannot observe is reported as Unable to Verify, never as a pass.

Agent Assurance capabilities at general availability

• Auto-generated test suites: Functional and adversarial scenarios derived from the agent’s docs, such as a PRD, policy docs or an API spec, and its endpoint. Adding the codebase is optional and gives deeper results.
• Red teaming: 25 attack patterns, including multi-turn attacks, mapped to all 20 risks in the OWASP Top 10 for LLM Applications and the OWASP Top 10 for Agentic Applications, and to eight MITRE ATLAS techniques. The report is organised risk by risk, showing what was tested, what each attack reached and what was verified.
• Evidence-backed verdicts: Every criterion ends as Pass, Fail or Unable to Verify, and each verdict quotes the evidence it rests on. Unable to Verify is never counted as a pass, and the report lists each unverified result separately with what is needed to verify it.
• Quality, performance and token economy: Beyond functional and adversarial tests, a run checks output quality and instruction-following, measures token economy (cost per task, verbosity and redundant tool calls) when the agent reports its token usage, and records latency for every turn.
• Root-cause analysis: On request, Rook groups failing tests and gives each group a likely cause and a suggested fix. Docs and an endpoint are enough to start; adding the agent’s code gives deeper root causes.
• Regression testing in CI: Each run is compared with the previous one, and tests that newly fail since the last run are flagged. Access-key sign-in works on CI runners with no browser, with recipes for GitHub Actions, Jenkins, and Argo CD.
• Multi-agent testing: Agent Assurance can grade a crew of agents that hand off work to each other with no human in the loop, or an n8n agent over its webhook.
• Controlled writes: Because an agent’s writes are real, Rook shows the run plan and asks for confirmation before an interactive run.
• No provider keys: Rook works out what the agent is for, writes the scenarios and runs the agent the way a user would, while the models run in TestMu AI’s controller. A hosted Cloud UI shows results for unlimited team seats.
• Audit-ready record: Every message sent to the agent under test, every reply, every tool call and every verdict is written to a session transcript, with credentials, tokens and keys masked.
• Works where developers work: Run Agent Assurance from the terminal or from inside Claude Code, Codex or Gemini CLI, on macOS, Linux or Windows, and install Rook with npm, Homebrew or the installer.

Speaking on the general availability of Agent Assurance, Mayank Bhola, Co-Founder and Head of Products, TestMu AI, said, "An agent can tell you it finished the job and still have done it wrong. Agent Assurance doesn't grade the reply. It grades what the run changed, and it says so plainly when it can't tell. That is why Unable to Verify is never counted as a pass. A pass rate padded with guesses is worse than no number at all, because teams ship on it. Early teams told us the hardest step was getting to a first run, so we removed the sales call. A team can now point Agent Assurance at its own agent on day one and see exactly how much of its behaviour has been checked."

"With an agent, the path can change on every run," said Vipul Verma, Group SVP Engineering, TestMu AI. "So we stopped trying to predict the path and started checking where the agent actually ended up: what it changed, what it sent, what it touched. With general availability, any team can point Agent Assurance at its own agent today and see what it actually did, not just what it says it did."

Availability

Agent Assurance is generally available today to anyone who signs up for a TestMu AI account. Users can start with a 14-day free trial that includes 4,000 credits and four concurrent runs. Scenarios and run results stay in the team’s repository as plain files, whether or not the team moves to a paid plan.

Agent Assurance is available at testmuai.com/agent-assurance.

Frequently asked questions

What is Agent Assurance?
Agent Assurance is testing that grades what an AI agent actually changed, not what it says it did. Rook CLI derives a suite from your code or docs, runs your agent against it, checks its tool calls and their effects through hooks the team connects, and reports separately what it could not verify. It is generally available and publicly installable.

Can Agent Assurance run in a CI/CD pipeline?
Yes. Agent Assurance supports access-key sign-in for headless CI runners and provides recipes for GitHub Actions, Jenkins, and Argo CD.

Which AI agents can TestMu AI test?
TestMu AI tests coding, workflow, back-office, and multi-agent systems with Agent Assurance, and chat, voice, phone, video, and image agents with Agent Testing.

Is there a free trial of Agent Assurance?
Yes, users get a 14-day trial with 4,000 credits and four concurrent runs.

About TestMu AI

TestMu AI (Formerly LambdaTest) is a full-stack agentic AI Quality Engineering platform that empowers teams to test intelligently and ship faster. Engineered for scale and AI-native by design, it offers end-to-end AI agents to plan, author, execute, and analyze software quality across web, mobile, and enterprise applications on real devices, real browsers, and custom real-world environments.

Fonte: Business Wire