FAQ

Short answers to the questions engineers, PMs, and QA ask most often when they first evaluate Cofactor. Each one links to the deeper reference.

01How is Cofactor different from Playwright, Cypress, or Selenium?

Traditional frameworks make you write and maintain code and CSS/XPath selectors, and that selector code is what breaks every time the UI shifts. With Cofactor you describe a journey in plain language and an AI agent drives a real browser, deciding which element to act on at run time. Under the hood it still drives Chromium, so you keep real-browser fidelity and just maintain intent instead of selectors. It complements an existing suite rather than forcing you to throw it away.

02Do I need to know how to code?

No. You describe what a test should do in plain language and the AI agent drives the browser, so there's no code to write and no selectors to maintain. Build a test visually in the Flow Designer, author it by chatting with the agent, or scaffold one from the CLI. Engineers, PMs, and QA can all create and run them.

03My UI changes constantly. Won't these tests be flaky?

This is exactly what the agentic model is built for. When a selector drifts or a page changes, Cofactor's agents re-resolve the element, update the cached locator, and keep the suite green without a human rewriting the test. That's self-healing. A flake classifier separates genuine intermittency from real regressions. A hardcoded selector breaks on the same change that an agent simply adapts to.

04How does login work, and how are credentials kept secure?

Add credentials on the property, then validate login with the preview tools before you kick off a large run. Credential templates give you one shared test identity per property with a different credential per environment, so the same flow signs in across staging, production, and sandboxes without edits. Credentials can be marked ephemeral, can pair a provisioned inbox for email sign-in codes, and can forward browser-based auth into API requests via Auth Handoff. See Credentials and Login.

05Can I test localhost, staging, or an app behind a VPN or corporate proxy?

Yes. If your staging environment is public, just add its URL as a property and you're testing in minutes. For private environments, connect your AWS account and Cofactor runs a managed Chromium browser inside your VPC, reaching your app over the internal network with an outbound-only encrypted WebSocket, so there's no inbound traffic and no public exposure. Behind a corporate firewall or proxy? Reach out to us and we'll help you get connected. See Testing Private Environments.

06How do I run tests in CI or on every release?

Install and authenticate the CLI in your runner, then trigger a single flow with cofactor workflow run <id> (or a whole suite) and stream the results back — the run exits non-zero on failure, so it gates the pipeline. See Using the CLI in CI and Running your tests.

07A test failed. How do I tell if it's a real bug or the agent getting confused?

Every run produces a deterministic trace: a full audit log of every navigation, action, assertion, and screenshot, plus the agent's own reasoning, so you can replay exactly what happened and see why. That audit log, paired with comprehensive AI diagnosis, is how you separate a genuine regression from an environment, data, or setup problem. Confirmed findings roll up into Issues with AI triage. For tests that fail cleanly and clearly, see Designing a good test.

08Where do my tests live and how do I run them?

Your tests are flows: visual, multi-step journeys stored in Cofactor. Create and edit them in the Flow Designer or by chatting with the agent, then validate, approve, and promote a flow when it is ready for suites. Run them from the dashboard, on a schedule, or from the CLI (cofactor workflow run <id>). See the CLI Quickstart.

09Can I use my own agent to build tests?

Yes! That's exactly what the CLI is built for. Coding agents like Claude Code and Cursor can drive Cofactor end to end: write a flow, run it (cofactor workflow run <id>), stream the results, and diagnose failures straight from the trace, all from the terminal. Every command is non-interactive and flag-driven, so your agent can work without a human at the keyboard. Run cofactor install to add the MCP server so your agent can call Cofactor directly, or cofactor skill <name> to drop in a ready-made agent skill. See Install the CLI and the command reference.

10How is usage measured?

Usage is measured in agentic runs: a run where the AI agent actually drives the browser. Platform safety limits (crawl/run caps, payload sizes) and request rate limits are documented under Limits and Rate Limiting. For plan and pricing specifics, talk to us.