What is Droidrun?

I've tested a lot of mobile automation tools. Most fall into two camps: brittle traditional UI automation or inefficient vision-based AI agents. Droidrun breaks that pattern entirely.

Built by a German team (Peter Lächner, Christian Ninstel, Niels Schmidt, Nikolai Dück, Timo Beckmann) and backed by €2.1M in pre-seed funding from Merantix Capital, Droidrun is the first native mobile AI agent that treats Android devices as first-class citizens. Instead of trying to "see" screens like a human would, it speaks the language of Android itself—accessibility APIs—giving it superhuman precision and speed.

The core insight is elegant: accessibility APIs already encode what's on screen in structured, machine-readable format. Every button, text field, and gesture has metadata. Traditional tools ignore this. Droidrun embraces it. The result is 91.4% success rate on AndroidWorld benchmarks—massively ahead of competitors and with execution costs at just $0.075 per task.

It's open-source (MIT license), LLM-agnostic (OpenAI, Claude, Gemini, DeepSeek, Ollama), and genuinely innovative. The cloud platform is still in development, but the framework is production-ready and actively evolving.

Key Features & Capabilities

Here's what makes Droidrun different from everything else I've tested:

How to Get Started

Getting Droidrun running is straightforward, though it requires some setup:

1. Install the Framework

Clone from GitHub. The MIT license means you own what you build. Documentation is solid, and the community is responsive.

2. Enable Accessibility

On your Android device, enable the Droidrun accessibility service. This is a one-time setup and gives the framework the hooks it needs to control the device natively.

3. Connect Your LLM

Plug in your API keys for Claude, OpenAI, or whichever model you prefer. Config is straightforward and well-documented.

4. Define Tasks

Write your automation in natural language or structured YAML. The planning engine handles complex multi-step workflows. You can also record macros visually and replay them.

5. Execute & Monitor

Run your tasks on your device or scale to multiple devices. The execution engine handles retries, error recovery, and real-time logging.

The learning curve is gentler than traditional automation frameworks because you're describing intent in natural language rather than writing brittle selectors. However, setup requires familiarity with Android and API keys.

Performance & Benchmarks

This is where Droidrun flexes. The numbers are genuinely impressive:

The speed advantage comes from the accessibility API approach. Instead of sending full screenshots to the LLM, Droidrun sends a compact, semantic representation of the UI. This means fewer tokens, faster decisions, and lower costs. Vision-based competitors process pixel data, which is information-dense but inefficient.

In practice: You get better results, faster execution, and lower costs. That's the triangle you rarely win on.

Pricing & Licensing

If you're budget-conscious, the open-source framework alone is production-ready. The incomplete cloud offering currently limits enterprise deployment, but the roadmap is clear.

Strengths & Weaknesses

My take: The strengths are genuine technical advantages. The weaknesses are mostly timing—it's a new project with clear roadmap. Android focus is actually a strength (better support) not a weakness if Android is your target platform.

How Droidrun Compares to Competitors

Here's how Droidrun stacks up against other mobile automation approaches:

Vision-based competitors (AppAgent, Mobile-Agent): Send screenshots to LLMs. This is intuitive but information-inefficient. Result: slower, more expensive, lower success rates.

Traditional automation (Appium, UIAutomator): Require hand-written selectors and brittle scripts. No AI planning. Work well for fixed UI but fail on variability.

Droidrun's advantage: Combines AI planning with native API control. Gets the best of both worlds. The accessibility API approach is genuinely novel and gives real performance advantages.

Team & Funding

Strong founding team with clear focus. Merantix Capital backing signals serious backing. The €2.1M pre-seed validates the market interest in better mobile automation. Community engagement on GitHub (8,000+ stars in first year) shows real adoption.

Final Verdict

Frequently Asked Questions