Agentiqa
Agentiqa is an agentic QA service for teams with no dedicated QA department. Point it at a URL: the agent explores the app in a real browser, plans tests, runs them, and reports bugs with evidence, without touching your code.
What is Agentiqa?
Agentiqa describes itself in its documentation as an AI QA engineer for web apps. You give it the address of an application; its agent opens that application in a real browser, works out what matters, runs test plans against it and reports the problems it finds with screenshots and reproduction steps. There is no repository access and no code upload: both the plan and the verification are derived from the rendered interface.
The working model is the agent proposes, you curate, with three checkpoints: the scope, where areas are ranked by risk; the test plan itself; and the findings. Only findings always require human review, scope and plan being approved by default, and in headless mode (CLI, CI, scheduled runs) all three checkpoints auto-approve.
Five execution surfaces are offered: the web app, whose browser runs in the cloud; the desktop app, an Electron build with a visible embedded browser; a CLI, with agentiqa explore and agentiqa run; coding-agent plugins for Claude Code, Cursor and Codex; and a GitHub Action.
A project is one URL under test plus everything the agent learns about it: the map, the findings, the plans and the memory. That project memory accumulates from one run to the next, carrying journeys, quirks and past bugs. The defects targeted are functional and UX regressions, broken states, visual defects and responsive issues, each with a severity, a category and reproduction steps.
Two guardrails sit around execution. An independent supervisor, a second model, watches every run and can let it continue, redirect it, block it or conclude it. And every run is recorded on video, so each bug leaves with its evidence attached. Runs are metered: one run equals 25 agent steps, and a run shorter than 25 steps still counts as one.
Two conveniences complete the testing surface. Each run gets its own test mailbox that the agent can read and reply from, which covers signup verification, magic links and email validation. And shared test accounts let several simultaneous runs use accounts of the same role inside a project. The publisher situates itself in real-UI AI testing, extended towards test generation.
What it does
- Point the tool at a URL, either the whole product or only the flows you have just shipped
- Let the agent explore the application in a real browser and map the journeys it finds
- Approve or redirect at three checkpoints: the scope, the plan, then the findings
- Receive bugs as structured reports with video, screenshots and step-by-step reproduction
- Replay a plan on demand, in parallel, or straight from your CI pipeline
- Turn the day's exploratory findings into a replayable regression suite
- Trigger a check from your terminal or from a coding agent such as Claude Code, Cursor or Codex
When to use Agentiqa / When not to
A quick filter to help you decide if Agentiqa is the right fit.
When to use Agentiqa
- Product teams shipping without a dedicated QA department, who want a QA hub they can simply point at a URL
- Teams whose interface changes faster than their tests are updated, in particular when coding agents regenerate components
- Engineering teams that want a verification pass independent from the agent that wrote the code
- Teams that need to test localhost and internal builds without uploading anything
- Teams wiring a quality gate into CI/CD, where GitHub Actions, GitLab, CircleCI and Buildkite are auto-detected, including unattended nightly regression runs of the kind Lio Technologies operates
When not to use Agentiqa
- Teams looking for unit testing, load testing or performance testing: the publisher states plainly that Agentiqa replaces none of them
- Teams evaluating LLM outputs, doing pure visual diffing, or testing APIs and components inside the application repository, all placed outside the scope
- Teams whose security review rules out letting a third-party AI touch pre-production builds, a case the publisher itself cites as a reason to stay on Playwright
- Teams that already have SDET capacity and a genuinely maintained suite, where the publisher admits the gain is small
- Teams that need strict determinism, with mocked network, seeded data and exact assertions, for which the publisher points back to Playwright
How to use Agentiqa
A typical end-to-end flow, from setup to results.
- Pick one of the three self-serve entry points, each announced as operational within minutes
- Lightest route, the web app: sign in from your browser, choose a project and describe what you want tested, with nothing to install and the browser running in the cloud
- Terminal route: run npx agentiqa, then drive the agent with the agentiqa explore and agentiqa run commands
- Desktop route: use the desktop app and its visible embedded browser to test localhost and internal builds (check availability first, see the watch-outs)
- For a first test, create a project with the URL of your application and ask the Assistant to test it
- Curate the three checkpoints in order: the scope, then the plan, then the findings
- Install the plugin for your coding agent, Claude Code, Cursor or Codex, to trigger checks where you already work
- Wire it into CI with project-scoped service keys, JSON output carrying a schema version, real exit codes and automatic detection of the CI provider
- Add the agentiqa/qa-action GitHub Action to replay saved plans and fail the job according to the result
- Tune parallelism from one to five simultaneous plans; on Starter and Pro there is no API key to manage, both being fully managed
Pros & Cons
Pros
- No repository access and no code upload are required: the verification is derived from the rendered interface
- The verification is independent from the coding agent that wrote the feature
- Credentials are encrypted with AES-256-GCM and never placed in the model context
- Every finding comes with verifiable evidence: video, screenshots and step-by-step reproduction
- Localhost and internal builds can be tested without a tunnel and without any upload
- It integrates natively with CI, with real exit codes and versioned JSON output
- A permanent free plan, Starter, offers 10 runs per month, plus a 14-day Pro trial
Cons
- The desktop app is sold in the plans and promoted throughout the site, yet its download page shows 'Coming soon' for macOS, Windows and Linux and serves unfilled placeholders ('Current version: v...', 'Released: ...'), as observed on 05/09/2026
- There is no public API and no API documentation: programmatic access goes through the CLI and the GitHub Action only
- A single customer is named publicly, Lio Technologies, so the field evidence is thin
- SOC 2 is not obtained, only in progress, and no other certification is claimed; Zero Data Retention exists on the Company plan alone, and the publisher specifies it is not a control automatically enforced by the product today
- No Article 27 EU representative is designated although the publisher declares itself controller for the EEA and the United Kingdom, and the site never says whether customer data is used to train models, nor offers any opt-out mechanism
- There is no intermediate tier between 39 USD per month and a quote-based Company contract
- The default model provider is imposed, Google Gemini, and bringing your own model is a Company-plan option; the site also contradicts itself, the security page naming Google Gemini alone while the privacy policy lists OpenAI, Anthropic and Google
Pricing & Plans
A free plan is available. Starter is free forever and includes a 14-day Pro trial, 10 runs per month and prepaid credit top-ups consumed as you go, with one run counted as 25 agent steps and any shorter run still counting as one. The lowest paid entry point is Pro at 39.00 USD per month. The Company plan is quoted individually as a volume-based contract, explicitly not per seat. Payments are handled by Lemon Squeezy as merchant of record, plan changes can be made at any time with immediate effect, and payments are non-refundable except where the law requires otherwise (terms, section 3). Prices as read on the pricing page on 05/09/2026.
- Agentiqa Desktop
- a 14-day Pro trial
- 10 runs per month and credit top-ups
- everything in Starter
- plus Agentiqa Web with cloud execution
- parallel runs and priority support
- tailored volume-based contract
- on-premises
- self-hosted or BYOK deployment
- guided setup and maintenance
- one-month warm start
- SLA and contractual Zero Data Retention
Data, GDPR & hosting
A consolidated view of how Agentiqa handles your data.
GDPR overview
GDPR implementation is concrete but incomplete. Agentiqa, Inc. declares itself data controller for people in the EEA and the United Kingdom, and cites its Article 6 legal bases: performance of a contract 6(1)(b), legitimate interest 6(1)(f), consent 6(1)(a) and legal obligation 6(1)(c). Rights are enumerated in full: access, rectification, erasure, restriction, portability, objection, withdrawal of consent and complaint to a local authority, with explicit pointers to the European Data Protection Board and the UK ICO. Transfers outside the EEA rely on standard contractual clauses and, where the provider is certified, the EU-U.S. Data Privacy Framework; a DPA is available on request. Californian rights are handled in a separate section. One gap is visible and must be stated: no Article 27 EU representative and no data protection officer are named.
Who owns the data?
Section 8 of the terms of service, effective 19 June 2026, leaves ownership with the customer. You keep every right over the applications, sites and content you test, and the test results, screenshots and reports the service generates belong to you: Agentiqa, Inc. claims no intellectual property right over your test data or your results. One reservation is written into the same section: the publisher may use anonymised and aggregated data derived from service usage to improve its products, and states that such data must identify neither you nor your tested applications. When the contract ends, test data remains exportable for a reasonable period of at least 30 days.
Reuse rights
The privacy policy, effective 19 June 2026, lists what is collected: account data (email, name, authentication), usage data, test data (URLs, screenshots, results), device data, payment data and support communications. It is used to deliver and improve the service, execute tests, personalise, communicate, process payments, detect fraud and analyse trends. The boundary the publisher draws around the model provider matters most: credentials are encrypted with AES-256-GCM, typed by name and never placed in the model context, and neither credentials nor source code are sent to the model provider. What does reach it is test execution context, meaning screenshots, structured page snapshots and your own test instructions. Cloud runs disable the file-reading tool; desktop runs keep everything on your machine. Payments go through Lemon Squeezy as merchant of record, no card number is stored by the publisher, and the publisher states it does not sell personal data.
Data retention & training
Hosting summary
Agentiqa, Inc. is established in the United States and declares that it processes data in the United States and in the other countries where its providers operate. The privacy policy names the infrastructure and hosting providers relied upon: Vercel, Google Cloud and E2B. Where the processing happens depends on the execution mode. In desktop execution, everything stays on the user's machine and only the calls to the model provider leave the network. In cloud execution, the browser is managed by the publisher, and test data is processed in the cloud and kept only for as long as it takes to deliver the result, unless it is explicitly saved. One statement of EU residency exists but is narrow: for the documented Lio engagement, the publisher indicates that execution and all stored evidence stay in the EU. That is a modality of that engagement, not a general guarantee of the product. Transfers are framed by standard contractual clauses and, where applicable, the EU-U.S. Data Privacy Framework.
Things to keep in mind
Risks and trade-offs to weigh before adopting Agentiqa.
- The desktop app is included in the Starter plan and promoted on the home page, the pricing page and the Starter page, but the download page shows 'Coming soon' for all three operating systems and displays no version number and no release date, serving raw placeholders in the HTML, as observed on 05/09/2026
- The usage figures published on the Starter page are dated: 219 versions shipped to npm and more than 15,000 npm downloads in June 2026. These are snapshots, not current values
- SOC 2 is in progress, not obtained, so it must not be read as a certification; and the Company plan's Zero Data Retention is a contractual commitment that the publisher declares is not automatically enforced by the product
- The site takes no position on whether models are trained on customer data, and documents no opt-out mechanism
- No Article 27 EU representative is designated while the publisher declares itself controller for the EEA and the United Kingdom
- The site contradicts itself twice: the security page names Google Gemini as sole model provider while the privacy policy lists OpenAI, Anthropic and Google, and a July 2026 blog post still publishes an outdated pricing grid that differs from the pricing page
- The 'Trusted by' block on the home page mixes one customer, Lio, with support programmes (Google for Startups, German Accelerator, TUM.ai E-Lab, Agentic Builders): those four are not customers
Setup & Integrations
Technical difficulty
Low. Getting started is announced without code and without scripts: you supply a URL. Three self-serve entry points are said to be operational within minutes, the lightest being the web app, which installs nothing; the CLI takes one command, npx agentiqa. Starter and Pro are fully managed, with no API key to handle and no repository integration required. The real entry cost is writing clear instructions: the publisher warns that natural-language instructions can be ambiguous. Wiring into CI is more technical, with service keys, exit codes and a JSON envelope. The Company plan adds one month of guided setup.
Deployment
Integrations
Behind Agentiqa
Social
Resources
All the official URLs gathered for verification and reference.
Alternatives
Tools that compete with or complement Agentiqa.
Frequently asked questions
What is agentic testing?
What kinds of problems does Agentiqa find?
Can I use it locally and in CI?
How are my code and my secrets protected?
How do credits work?
Do I need to bring my own API key?
Is there a public API?
Which model provider is used?
Is Agentiqa SOC 2 certified?
How is this different from plugging an AI into a browser?
Should you pick Agentiqa?
Agentiqa arrives with a clear and defensible position: an independent verification pass, run by an agent that did not write the code, driven from nothing more than a URL. The company is declared founded on 11/05/2026, Agentiqa Web was announced on 28/02/2026 and the npm package has been published since March 2026, so the product is young and its public track record is correspondingly thin, with a single named and quoted customer reference, Lio Technologies.
What sets the publisher apart is an unusual willingness to document its own limits. It writes that SOC 2 is in progress rather than obtained, that the Company plan's Zero Data Retention is a contractual policy and not a control enforced by the product today, and that vision-based selection can drift on unusual layouts while natural-language instructions can be ambiguous. That candour is worth more than a page of claims.
The main reservation is a contradiction the site leaves open. The desktop app is central to the pitch and is a headline item of the free Starter plan, yet its download page still showed 'Coming soon' for macOS, Windows and Linux on 05/09/2026. Anyone counting on local testing of localhost and internal builds should confirm availability before committing. A secondary reservation is regulatory: nothing is published about whether customer data trains models, no opt-out exists, and no Article 27 EU representative is named while the publisher declares itself controller for the EEA and the United Kingdom.
For a team shipping fast without a QA department, the free plan and the CLI make trying Agentiqa almost costless, which is the right way to judge it. For a team with an SDET function and a maintained suite, the publisher itself says the gain will be small.
- Choosing a selection results in a full page refresh.
- Opens in a new window.