Your AI. Your keys. Your boundary.

Four providers, all of them yours

Paste a key. That's the integration. The AI call happens between WPLoadTester and the provider of your choice. WPI never sits in the middle, never sees your prompts, never stores your data.

We test the models, so you don't have to guess

Bring-your-own-AI only pays off if the model can actually do the work. We benchmark the AI Assistant against the auth and correlation patterns that break recordings in the real world: PKCE and OAuth logins, SAML federation, and JavaScript bot challenges. Then we tell you which models clear them from start to finish, so you can point your team at one that finishes the job instead of one that stalls halfway.

Smaller or older models are fine for simple recordings and quick questions, and you are always free to use any model your provider offers. For the toughest correlation work, though, a validated model is the difference between a recording that replays clean and one that almost does.

What the AI sees once you've configured it

Once you've pasted your key, the AI Assistant lives inside WPLoadTester, analyzing your test cases, identifying auth patterns, and explaining what it's about to do before it does it.

Underneath the AI is a deterministic expert system battle-tested against thousands of real test cases. The rules for ASM correlation, dynamic-field extraction, and session-state recognition aren't being invented by the LLM each time. The AI decides which rules to apply and explains why; the expert system handles the precision work. That's why auto-configuration works on real-world test cases with hundreds of dynamic fields, instead of breaking on the third dynamic variable like a pure-LLM tool would. See the diagnostic playbook for the symptom taxonomy the AI consults when it analyzes results.

And it gets cheaper the more you use it. The rules the AI creates don't disappear when the session ends. Once it has worked out how to correlate a session token or extract a dynamic field on your application, the expert system keeps that rule. The next time you configure a test case against the same site, the expert system applies those saved rules automatically, so the AI isn't asked to solve the same configuration problems again. The system learns how your application behaves from past runs, which means less repeated AI work and lower token spend on top of the no-markup pricing.

When the AI hits something tricky, you see it think.

For harder configuration problems (extracting a JWT from a streaming Next.js payload, correlating dynamic variables across pages, debugging a regex that isn't matching), the AI walks through its attempts out loud instead of silently giving up. You see exactly what it tried, what it observed, and why it's refining.

And after the test runs, the AI interprets the results.

WPLoadTester 7's Load Test Analytics dashboard produces AI-written views of every run: a plain-English performance narrative, and a bottleneck analysis that picks out inflection points and traces root causes from the actual numbers.

Bring your own agentic toolkit via MCP

WPLoadTester 7 also exposes its test cases, results, and AI workflow as an MCP (Model Context Protocol) server. Point any MCP-compatible agent at it (Claude Code, Cursor, your own CLI tool, anything you build) and drive the whole product from your terminal. The AI layer isn't a black box; it's extensible end-to-end.

This is the senior-engineer / agentic-CLI surface. If you'd rather work from a chat panel inside the app, see the in-app AI Assistant described above.

Read the full MCP server feature page →

The Azure angle: Microsoft-first enterprises

If your company already runs on Microsoft (Entra ID, Azure subscriptions, an enterprise agreement), Azure OpenAI is the provider your team will approve fastest. The GPT models run in your own Azure OpenAI resource, in the region you choose, inside the subscription your security team already governs.

Microsoft's commitment is explicit: your prompts and completions are not used to train its models, and are not shared with OpenAI. Private networking, customer-managed encryption keys, and your existing compliance certifications (SOC 2, HIPAA, ISO 27001) carry over because the resource is yours, not a new vendor's. It's the same data-residency story as Bedrock, told in Azure terms.

The Bedrock angle: regulated and air-gapped buyers

If your organization is regulated (financial services, healthcare, federal) or operates in isolated AWS environments (GovCloud, dedicated regions), AWS Bedrock support is the wedge.

Bedrock runs inside your AWS account, subject to your IAM policies, within your compliance boundary. If Bedrock is already approved for your workloads, WPLoadTester using it for AI auto-configuration is the same data-residency story, not a new one. GovCloud regions and FedRAMP-authorized environments work identically.

If Bedrock isn't an option (fully air-gapped, no AWS), self-hosted AI is on the roadmap. Ollama, vLLM, and any OpenAI-compatible endpoint will plug into the same provider-paste workflow, just pointed at your own model server. Talk to us if that's blocking your evaluation.