AI Studio

AI Studio is Robonito's autonomous testing agent. Instead of recording clicks manually, you describe what you want to test in plain language, and the AI agent browses your application, generates test steps, saves them as real test cases, and even classifies failures — all without writing a single line of code.

AI Studio main interface showing the chat panel, browser mirror, and platform selector

Start here

I want to…	Go to
Run my first AI-generated test	Quick start →
Understand the key terms (thread, suite, persona, HITL, recipe)	Concepts →
Run the web agent from a terminal or call the API	Advanced — CLI and API →

This page is an overview. The deeper, task-based guides live in the sections below and in the sidebar.

What AI Studio Can Do

Generate test cases from a natural language description — describe a user flow and the agent executes it on a live browser, capturing each step automatically.
Explore your application autonomously — the agent crawls your entire app surface using a breadth-first strategy and maps all discovered pages and interactions.
Plan before acting — in Plan Mode, the agent researches your goal and presents a step-by-step strategy for your approval before touching the browser.
Classify bugs — when a test run fails, the agent analyzes the failure and labels it as a real bug, environment issue, broken selector, or flaky test.
Save directly to your test suite — generated test cases are one click away from being added to any suite in your workspace.
Support multiple AI providers — use Robonito's default model or bring your own API key for Anthropic Claude, OpenAI GPT, Google Gemini, OpenRouter, NVIDIA NIM, or a local Ollama model.

Supported Platforms

AI Studio works across all test platforms Robonito supports:

Platform	Description
Web	Automates a Chromium browser against any web application
API	Generates API test cases from endpoint descriptions or Postman collections
Android	Controls an Android emulator via ADB — requires Mobile Setup
iOS	Controls an iOS simulator via Appium — requires macOS with Apple Silicon
Desktop	Automates desktop application workflows

Agent Modes

The mode sets how the agent approaches your goal. It appears in the input toolbar for the Web platform.

Default Mode

The agent immediately begins interacting with the application, translating your natural language prompt into browser actions, validating results, and generating automated test steps as it goes.

Plan Mode

The agent first analyzes your goal and produces a Strategic Plan — an interactive step-by-step breakdown of its intended approach. You review, refine, and approve the plan before the agent touches the browser. Once approved, the agent executes the plan and presents a review card to save the resulting test case.

To activate: click the Plan toggle in the input toolbar before sending your message.

Plan Mode approval card showing the strategic plan before execution

Explore Mode (Web only)

The agent autonomously explores your application in resumable segments:

Walks real end-to-end user journeys (signup, search, checkout) and saves distinct tests directly to your suite.
Employs an Explore Save Guard with suite-wide behaviour deduplication and value gating so you don't accumulate duplicate or trivial tests.
Automatically captures the initial login flow as an Auth Test and replaces repetitive downstream logins with lightweight REUSE_TEST steps.
Generates synthetic form data dynamically with unmasked test values.
Discovers broken business rules (e.g. incorrect price totals or invalid state transitions) and automatically logs them as bugs while saving failing regression checks.
Validates backend server rules using the browser's active session (BROWSER_SESSION).

To activate: click the Explore toggle in the input toolbar.

Explore Mode Site Map tab showing a visual graph of discovered pages

Plan and Explore are mutually exclusive — choose one per run. See Choose an agent mode.

Configure the AI Model

Click the model selector dropdown to choose the provider and model:

Robonito (Default) — fully managed, no API key needed.
Anthropic Claude — Claude Sonnet 4.6, Opus 4.6, or Haiku 4.5.
OpenAI GPT — GPT-5.5, GPT-5.4 Mini, or GPT-4.1.
Google Gemini — Gemini 3.1 Pro, 3 Flash, 2.5 Flash, or 3.1 Flash Lite.
OpenRouter — browse hundreds of models from a single API key.
NVIDIA NIM — browse NVIDIA's NIM model catalog.
Ollama (Local) — run any locally installed Ollama model on your machine.

Model selector dropdown showing all available AI providers

Managed Keys: Enterprise provider keys are pre-configured by the Robonito team on your workspace during onboarding. Custom Bring-Your-Own-Key (BYOK) configurations can also be managed in Settings → External APIs. See Configure a provider and model.

Test as a Persona

The agent can adopt a persona — a real user of the application under test (a clinician, a call-center agent, a sales rep, …) — and exercise your app the way that person would, while still applying QA discipline. Ten personas ship by default, and you can add your own.

See Test as a persona and Persona testing — demo sites.

Attaching Context

You can give the agent richer context before it starts:

Files — attach text files, images (screenshots/mockups), or documents using the Attach (paperclip) button.
Figma frames — type /figma in the chat to link a Figma frame URL. The agent uses the design as a reference for what the UI should look like.
GitHub files — type /github to link a specific file or PR for context.
Jira issues — type /jira to attach a Jira issue. The agent reads the acceptance criteria and uses it to design the test.
Linear issues — type /linear to link a Linear issue.
Asana tasks — type /asana to link an Asana task.

Slash-command menu in the chat input showing /figma, /github, /jira, /linear, /asana options

Human-in-the-Loop (HITL)

During a complex flow, the agent may pause and ask for your help — for example, when it encounters a CAPTCHA, a 2FA prompt it cannot solve, or an ambiguous UI state. A HITL card appears in the chat asking you to take over the browser momentarily. Once you complete the blocked step, choose whether to teach the agent the action, resume without teaching, or reject — and the agent continues or ends accordingly.

Human-in-the-loop takeover card with Take Over and Resume buttons

See Human-in-the-loop.

Add a Testing Focus

For Web platform, opt in to specialist lanes before the run:

Responsive — check the page across multiple device viewports.
Accessibility — automated accessibility conformance.
Performance — Core Web Vitals plus network metrics.

See Add a testing focus.

Saving Generated Test Cases

When the agent finishes, a test case card appears inline in the chat showing the generated steps. To save it:

Click Save to Suite on the test case card.
Select the target suite from the dropdown.
Click Save. The card shows a ✓ Saved indicator and the test case is immediately available in your Test Explorer.

Generated test case card with Save to Suite button and suite selector dropdown

Session History, Rewind & Fork

Every AI Studio conversation is saved as a Thread. To revisit a past session:

Click the History icon (clock icon) in the AI Studio header.
Browse your thread history — each entry shows the original prompt and the platform used.
Click any thread to resume it. The agent loads the full conversation context and can continue from where you left off.

You can also rewind a thread to an earlier point, or fork it into a new conversation.

History drawer showing a list of past AI Studio threads

See Thread history, rewind & fork.

Bug Classification

After an AI-generated test is saved and executed, if a step fails, AI Studio classifies the failure:

Classification	Meaning
`real_bug`	A genuine defect in the application
`env_issue`	The environment was misconfigured or unavailable
`selector_broken`	The UI element's selector changed since recording
`flaky`	The step passes sometimes and fails other times
`inconclusive`	Not enough evidence to classify

Classifications are shown as colored pills in the chat alongside severity (low, medium, high). Use the Log Issue button on any bug pill to create a Jira, Linear, or Asana issue directly from the failure.

Bug classification pill showing type, severity, and Log Issue button

Token Usage

A Token Tracker in the AI Studio header shows real-time token consumption for the current session — input tokens, output tokens, and tokens-per-minute. This is especially useful when using models billed by the token (Anthropic, OpenAI, Google). The tracker resets at the start of each new thread.

Mobile Testing with AI Studio

For Android and iOS platforms, the agent mirrors the device screen live and interacts via touch events (tap, swipe, type) just like a human tester. Select the target device from the Device dropdown when using Android or iOS mode.

See also: Mobile Setup →

All AI Studio docs

Topic	Page
First run, end to end	Quick start
Key terms explained	Concepts
Default / Plan / Explore	Choose an agent mode
Test as a user	Personas
Answer questions / take over	Human-in-the-loop
Responsive / a11y / performance	Testing focus
Pick a provider & model	Provider & model
Rewind / fork a thread	Thread history
Slash commands, models, flags, API	Reference
Terminal & app-server API	Advanced — CLI and API

Teams

Quick start