Exploratory tests can run on your CI pipeline just as unit and e2e tests do. Everything stays local, no external access from cloud services.
Verify a single function or component in isolation.
Drive the whole app through fixed user journeys.
A curious AI agent that investigates with intent.
Explorbot makes the third layer as routine as the first two.
Point Explorbot at a running web app with an isolated workspace inside it, so it can explore freely without touching real data.
The Playwright library with its browsers. The same engine your end-to-end suite already runs on.
Access to fast AI models from any provider. We recommend OpenRouter for the best speed and choice.
Nothing else. No GPUs, no virtualization and no special runners. If your CI runs Playwright, it runs Explorbot.
in tokens (no extra charge)
AI models that run at over 200 tokens per second:
OpenRouter Using OpenRouter Recommended first choice for Explorbot. Best price & speed models selection. OpenRouter setup with recommended models and providers costs under $1 per hour. OpenAI Platform Using the OpenAI Platform Run gpt-luna for agentic control and gpt-nano for raw interaction with pages. OpenAI setup costs approximately 1.5$ per hour.
A real session: the agent plans, drives the browser and writes tests on the fly — no human in the loop.
Your Playwright and CodeceptJS suites re-run the same fixed steps every build. Explorbot re-explores the very same pages new ways — clicking new UI elements and paths your scripts never touch.
Regression suite
Same steps · every build
Explorbot
New ways & new UI · every run
Scripted tests repeat the exact same steps every build. Run Explorbot beside them and it re-explores the same pages new ways each time — clicking new UI elements and paths your script never touches — to surface regressions hiding just off the scripted route.
A brand-new feature with zero automated coverage? Point Explorbot at it. From the live UI plus the specs and domain knowledge you give it, it works out the basic test cases and runs them straight away.
Point Explorbot at the app you already run. Here is how to set it up — and the kind of glitches it surfaces that scripted suites walk right past.
Add credentials, then point it at its main tables, forms and settings.
Catches glitches like
Add credentials, then point it at the deals pipeline and record pages.
Catches glitches like
Add credentials, then point it at the orders and inventory modules.
Catches glitches like
Explorbot never reads your source. It scans the live page, finds every interactive element and indexes it as a locator.
Explorbot expands dropdowns and collapsable elements to discover what is hidden.
Give Explorbot a goal and a URL — a team of specialised agents takes it from there, with no scripts and no human in the loop.
Maps the UI into sections and indexes every element — no docs or source needed.
Drafts test scenarios across normal, curious and edge styles — state changes first.
Drives the browser step by step; the Pilot breaks dead loops and adapts as the app changes.
Verifies each outcome before passing, clusters findings by root cause and captures evidence.
Saves passing flows as Playwright tests with reports and screencasts — and learns for next run.
Cheap, fast workers do the clicking and reading; smart managers make the calls — so a full session costs cents, not dollars.
Maps each page into sections, locators and an element index.
Drafts scenarios and cycles testing styles: happy path vs. negative cases.
Drives the browser one real step at a time to run each scenario.
Supervises the Tester and breaks it out of dead loops.
Orchestrates the session and takes your commands in real time.
Executes browser actions with resilient locator fallbacks.
Writes the run report, clustering defects by root cause.
Saves successful tests as markdown or Playwright or CodeceptJS code.
Seeds data via REST API using OpenAPI spec or learns from intercepted XHR requests.
Every run is summarized into a report your whole team can read: a clear pass and fail breakdown, the pages and features it covered, and a written analysis of what works and what broke.
Shown here in Testomat.io. Self-hosted runs also produce standalone HTML and Markdown reports you can open or commit anywhere.
Also produces
Playwright or CodeceptJS specs for every flow.
Step-by-step screencasts of every run as evidence.
Knowledge it reuses to test smarter next run.
ApiBot and DocBot bring the same autonomy to your APIs and your documentation.
ApiBot autonomously tests your API endpoints — hunting for security breaches, edge cases, and the request combinations a scripted suite would never try.
DocBot walks through your app and builds documentation straight from the live UI, so it is never outdated. It is not a spec — it captures how things actually look and behave right now, even when no spec exists or the UI has grown past it. It documents the real, current state of your app.
Set up Explorbot manually or make an agent do the job for you.
Add Explorbot to your project, then scaffold a config and a
.env
file with a single command.
npm i explorbot npx explorbot init
Add the skills to your agent, then run
explorbot-setup.
It installs Explorbot, writes the config, and proves it can reach
your app before handing back to you.
npx skills add testomatio/skills claude /explorbot-setup Install it, read the full documentation, and run it on your own app — it's all on GitHub.
Your data, the models, and how Explorbot fits next to the tools you already use.
Yes. Explorbot runs locally and keeps everything in files on your own machine. It never ships your data to third-party services, with a single exception: the AI provider you configure. If you trust that provider, you can trust Explorbot. Optionally, it can send traces to Langfuse for logs and traceability, but you can run Langfuse inside your own infrastructure. Explorbot is built to run within your own setup, including CI, where it reaches internal services with no external connections beyond the AI endpoint.
We recommend pointing Explorbot at a local, dev or staging environment rather than production. It can create, change or delete records through the UI, so it belongs in an isolated workspace seeded with disposable data, signed in as a user with limited permissions and no real secrets in reach. You can still run it on production for regression testing, as long as that account is scoped to a single project or workspace and cannot touch anything critical.
Explorbot drives fast, cheap models over pay-per-token APIs such as OpenRouter or Groq. You cannot run it on a coding-assistant subscription like Codex, Claude, Gemini, or Copilot. Those plans only expose models that are slow and costly, which are not suited to the way Explorbot reads large amounts of HTML on every step. Pay-per-token open models keep a full run at roughly a dollar an hour.
Yes, through the OpenAI Platform. Add an API key from platform.openai.com and point Explorbot at gpt-nano for reading pages, and gpt-luna for agentic decisions and vision. This is API access billed per token. Pricing should be as low as 1.5$ per hour of extensive work.
Cursor, Codex, and Claude Code are coding agents. They help a developer write tests interactively. Explorbot is built to run on its own for hours at a time, on CI, with no human in the loop. It is designed for continuous testing, not pair programming with a developer.
Yes. If you host your own models with vLLM, Ollama, AWS Bedrock, Azure, you can point Explorbot at any inference endpoint through the Vercel AI SDK. Speed and pricing then depend entirely on your infrastructure. For most teams, outsourcing to OpenRouter or Groq is simpler and more cost-effective than maintaining dedicated GPU capacity.
Explorbot fits CRUD-heavy web apps best: SaaS platforms, ecommerce, ERP, admin panels and internal tools. It is not a good match for landing pages, blogs or static sites. It works best when every screen has its own URL and uses titles and headings consistently, so it can tell one state from another. Point it at an isolated workspace or staging environment seeded with non-critical sample data, using an account with limited permissions and no real secrets.
Read the full prerequisitesExplorbot is source-available under the Elastic License 2.0 (ELv2), the same license used by Elastic and Grafana. It is free for commercial use: you can test any application, including commercial products, modify the code, self-host it and build it into your workflow at no cost. The one restriction is that you may not resell Explorbot itself as a hosted or managed service.