Tasker – Open-Source AI Browser Automation

TL;DR

TL;DR: Tasker is a free, MIT-licensed desktop app that records browser workflows or accepts plain-English descriptions, then executes them using AI (Claude, GPT-4o, or Gemini) with screenshot-based page understanding.

What Is Tasker?

Tasker is an open-source desktop agent for browser and OS automation, built by schnetzlerjoe. Unlike traditional RPA tools that replay exact coordinates and selectors, Tasker captures your actions as semantic hints — it knows you “clicked a Submit button” rather than “clicked at coordinates (234, 567).” When the page layout changes or a popup appears, the AI adapts rather than breaking.

The project is hosted on GitHub under an MIT license.

Setup Workflow

Step 1: Download and Install

Tasker is available for Windows, macOS, and Linux. Download the installer from the official site:

# Visit the official downloads page
open https://automatewithtasker.com/

Pre-built binaries are provided. On macOS, download the .dmg and drag Tasker to Applications. On Linux, use the AppImage or .deb package.

Step 2: Configure Your AI Provider

Tasker requires an API key from at least one supported LLM provider. Open Settings and add your key:

| Provider | Example Models | |----------|---------------| | Anthropic | Claude Opus 4.5, Claude Sonnet 4.5, Claude Haiku 4.5 | | OpenAI | GPT-4o, GPT-4o Mini | | Google | Gemini 2.5 Flash, Gemini 2.5 Pro, Gemini 3 Flash, Gemini 3 Pro |

Only models with configured API keys will appear as execution options.

Step 3: Record Your First Workflow

There are two ways to create a workflow:

Recording mode: Click Record and perform the task in the browser. Every click, scroll, and input is captured as a semantic hint rather than a raw coordinate.

Plain-English mode: Click New Workflow and describe what you want in natural language. The AI generates the execution plan from your description.

Step 4: Run the Workflow

When you run a workflow, Tasker:

1. Takes screenshots of each page step
2. Feeds the DOM structure + screenshot + your recorded hints to the AI
3. The AI decides the next action based on actual page content
4. Repeats until the stop condition is met

You can run in headed mode (watch the browser) or headless mode (runs in background).

Deeper Analysis

AI-First vs. Macro Playback

Traditional RPA tools like Selenium IDE, UiPath, or AutoIt replay exact actions. They fail when:

• A button moves 10 pixels
• A popup appears mid-flow
• The page structure changes between runs

Tasker sidesteps this by feeding the AI real screenshots and DOM context at every step. When a page element shifts, the AI still recognizes the “Submit button” even if its position changes.

Execution Control

Each workflow has configurable limits:

execution:
  mode: ai_assisted
  llm:
    provider: anthropic
    model: claude-sonnet-4-5

limits:
  timeout_seconds: 300
  max_steps: 50

You can also set stop conditions — for example, “stop when you have collected 10 results” — so the workflow terminates automatically.

Portability

Workflows export as portable YAML Taskfiles:

taskfile: "1.0"
metadata:
  name: "Search for products"
  description: "Search an e-commerce site and extract results"
variables:
  - name: search_term
    type: string
    required: true
  - name: max_results
    type: number
    default: 10

Share Taskfiles with teammates or import community workflows without reconfiguring from scratch.

Headless and Iterations

Run workflows headless for server-side or CI-style tasks. You can also set iteration counts (1–100) to run the same workflow against different inputs.

Practical Evaluation Checklist

• [ ] Downloaded and installed the desktop app
• [ ] Configured at least one LLM provider API key
• [ ] Recorded a simple 3-step browser workflow
• [ ] Ran the workflow in headed mode and verified AI adaptation on a layout change
• [ ] Tested headless mode for a background task
• [ ] Exported a workflow as a Taskfile YAML and re-imported it
• [ ] Verified stop conditions work (e.g., max results reached)
• [ ] Reviewed run history and screenshots for a multi-step run

Security Notes

• API keys are stored locally in the app’s settings, not sent to any external server beyond the LLM provider.
• Data stays local — Tasker runs entirely on your machine. Your browser data and workflow inputs do not go to a third-party service (beyond your chosen LLM API).
• Open-source code — the MIT license means you can audit the source on GitHub for any security concerns before using it with sensitive credentials.
• Headless mode caution — when running headless with API keys configured, ensure your machine is appropriately secured (screen lock, firewall) since the browser runs without a visible window.

FAQ

Q: Does Tasker work with any browser? A: Tasker uses a built-in browser ( Tauri’s webview, Chromium-based on most platforms). It does not control an existing Chrome/Firefox installation directly.

Q: Can I use Tasker without an API key? A: No. Tasker requires an API key for at least one supported LLM provider (Anthropic, OpenAI, or Google) to execute workflows with AI adaptation.

Q: How is this different from Puppeteer or Playwright? A: Puppeteer and Playwright are code-based browser automation libraries. You write JavaScript or Python to control the browser. Tasker is a no-code desktop app where you record clicks or write plain-English descriptions, and the AI handles the execution logic.

Q: Does Tasker work on Linux? A: Yes. Linux is a first-class platform with AppImage and .deb packages available alongside the Windows and macOS builds.

Q: Is there a self-hosted option? A: Tasker is a desktop app, not a server tool. There is no official self-hosted or Docker deployment option. For server-side browser automation, consider tools like Playwright, Puppeteer, or DrissionPage.

Conclusion

Tasker fills a specific gap in the browser automation space: between brittle macro recorders and full code-based frameworks like Playwright. By capturing actions as semantic hints and using AI to interpret page state at each step, it handles layout drift and popups that would break traditional RPA tools — without requiring you to write a single line of code.

For power users who want plain-English workflow creation, MIT-licensed transparency, and multi-provider AI flexibility (Claude, GPT-4, Gemini), Tasker is worth a look. The main trade-off is that it requires API keys for a supported LLM provider, and it is a desktop app rather than a self-hosted or CLI-first tool.

If you need a server-side or CI-friendly automation framework, stick with Playwright or DrissionPage. If you want a free, no-code way to automate multi-step browser workflows on your local machine, Tasker gets you there.

Source and Accuracy Notes

• Project page: automatewithtasker.com
• Source repository: github.com/pitalco/tasker
• License: MIT (verified via GitHub API)
• HN launch thread: news.ycombinator.com/item?id=46457912 (33 points, 29 comments)
• Source last checked: 2026-06-30 (README commit main)