Tanvrit
Automator

Test everything. Touch nothing.

An autonomous AI-powered QA agent for macOS that drives a real browser, understands your UI, fills forms with realistic test data, and keeps testing until every flow is verified — all without writing a single test script.

  • Autonomous multi-step flow testing

  • AI-powered element detection & interaction

  • Realistic test data via Faker — no hard-coded values

  • Learns flows and replays them without LLM calls

💻 Desktop

Automator shares your data with the rest of the Tanvrit platform.

The QA Agent That Never Gets Tired

Manual QA is expensive, slow, and inconsistent. Test scripts are brittle — one UI change breaks fifty tests. Tanvrit Automator solves this with a different approach: a fully autonomous AI agent that understands your website the same way a human tester would, but runs 24/7, never misses a step, and never files a Jira ticket when it's too tired to test properly.

Observe → Decide → Act → Learn

Automator runs a continuous loop: it observes every element on the page using Playwright's DOM API, asks a local LLM what action makes most sense toward your goal, executes that action in a real Chromium browser, then evaluates whether progress was made. No selectors to maintain. No test scripts to write. No mocking — it tests your actual live site or staging environment exactly as a real user would.

Realistic Data, Every Run

Automator detects what each form field expects — email, phone number, full name, product description, date — and generates realistic data automatically using Kotlin Faker. No more “test@test.com” polluting your database. No more hardcoded values that fail validation. Every test run uses fresh, plausible data that exercises your validation logic properly.

Local AI — Zero Cloud Costs

Automator connects to Ollama running on your own machine. There are no API keys, no per-test charges, and no data leaving your network. Run it against internal staging environments that are never exposed to the internet. The entire stack — browser, AI, database — lives on your Mac.

Flow Memory and Instant Replay

Every successful test flow is recorded to a local SQLite database. On subsequent runs, Automator recognises the URL pattern and replays the saved flow directly — no LLM calls, no uncertainty, just instant deterministic execution. It builds up a library of verified flows over time, giving you both exploratory AI testing and fast regression replay in the same tool.

Frequently Asked Questions

Does Automator require writing test scripts?
No. You provide a URL and a plain-English goal — Automator figures out what to click, type, and verify on its own using AI-powered perception and decision-making.
What LLM does Automator use?
Automator uses Ollama running locally on your machine — no cloud API keys, no per-request costs. Any Ollama-compatible model works, including Mistral and Llama 3.
How does Automator detect UI elements?
Automator uses a dual perception system: DOM extraction via Playwright for structured pages, and OCR-based vision fallback for canvas or WASM-rendered content.
Does Automator remember flows between sessions?
Yes — successful test flows are saved to a local SQLite database. Automator can replay them on demand without making any LLM calls, making reruns instant and free.
Is Automator available on Windows or Linux?
The current release targets macOS (Apple Silicon and Intel). Windows and Linux builds are planned for future releases.