---
title: "mobilerun · cloud phones for AI agents"
description: "Cloud phones for AI agents. Real Android & iOS devices in the cloud, plug in Claude Code or your own AI. Network, SIM, API, device, all from code."
url: "https://mobilerun.ai/"
---

# mobilerun

Cloud phones for AI agents. Start on the free plan (paid plans from $20/month), add a dedicated Android or a real iPhone, or connect your own phone, and drive it over an API with real device and agent identity, residential network routing and an optional phone number. Built on the open-source mobilerun framework (MIT), with published benchmark results: 98.34% mean official score across 133 iOSWorld tasks; 116 successful AndroidWorld trajectories.

mobilerun runs cloud-hosted Android and iOS devices, and
gives an AI agent everything a person's phone has: a consistent device identity,
a network route that matches it, an eSIM or a rented receive-only number for SMS codes, a permanent API inbox for
email, and state that survives between sessions. The open-source mobilerun framework is the engine
underneath it, MIT licensed.

Start local with `pip install mobilerun` (Python 3.11 to 3.13, Android 10 and later) on a device you already own, then move
to hosted devices with the Cloud SDK (`@mobilerun/sdk`, `mobilerun-sdk`).

## Key facts

- Pricing: a free plan to start; paid plans from $20/month, billed monthly, with extra usage at $0.03 per device minute. Dedicated Android from $64/month billed annually ($80 billed monthly). Up to 100 concurrent phones self-serve, unlimited on Enterprise.
- Enterprise: unlimited concurrent phones, outbound SMS, a custom SLA and individually agreed prices, plus private fleets, custom networking (private or bring your own) and eSIM provisioning, with iOS and Android in your regions behind your access controls. Book a call: https://cal.com/christian-ninstel-bvwhql/15min
- Devices: hosted Android on every plan, dedicated Android and real, physical iPhone bundles with a dedicated residential IP included, or your own Android or iPhone connected from the cloud. A stopped phone keeps its apps and data and bills no device minutes.
- Identity and network: one consistent device identity per phone (model, serial, IMEI, Android ID, advertising ID and carrier). The agent taps, types and swipes like a person. mobilerun Connect is our own network: residential traffic in every plan, dedicated residential IPs, or bring your own proxy. A dedicated IP keeps one consistent, private session between the phone and its exit, in 21 countries: Australia, Brazil, France, Germany, Hong Kong, Hungary, India, Indonesia, Italy, Japan, Malaysia, Philippines, Singapore, South Africa, South Korea, Spain, Taiwan, Thailand, United Kingdom, United States and Vietnam. Mobile carrier IPs are available on Enterprise, under contract: the pool is smaller than residential, so we scope it with you first.
- SMS: Receive SMS on any plan with a receive-only number (from $60/mo) or your own eSIM. Sending SMS is available on Enterprise.
- Benchmark results, record-breaking and with every run public: a 98.34% mean official score across 133 iOSWorld tasks (123 full, 10 partial), and 116 successful AndroidWorld trajectories: every one of the 116 AndroidWorld tasks received an official score of 1.0 (100% official score). 249 task replays and 8,906 recorded interactions: https://mobilerun.ai/benchmark/
- Benchmark setup: evaluated in September and October 2026 with mobile-harness, app cards and shared memory, 30 minutes per session: AndroidWorld with gpt-6-astra (low effort) on a Pixel 6 emulator (Android 13, API 33) with the official evaluation, iOSWorld with gpt-6-luna / high (98 tasks); gpt-6-astra / medium (35 tasks) on an iPhone 17 Pro simulator (iOS 26.5) scored by GPT-5.4 mini (119 tasks); GPT-6.1 Sol medium (14 tasks). The benchmark devices were an emulator and a simulator, not hosted mobilerun phones. Methodology: https://mobilerun.ai/benchmark/method/
- Benchmark comparison: the public iOSWorld leaderboard (https://iosworld.io/, checked September 2026) reports pass rates on the same 133 tasks, with a GPT-5.4 Mini judge and at most 50 interaction steps per task. It is led by Claude Opus 4.6 (vision+XML) at 51.9%, then Claude Sonnet 4.6 at 46.6% and GPT-5.4 at 39.8% (both vision+XML). mobilerun's run is not a leaderboard submission and was set up differently: mobile-harness with app cards and shared memory, a session budget of 30 minutes and corrected data fixtures for 3 tasks. Its binary pass rate is 92.48%: 123 of 133 tasks with a full score of 1.0; the 98.34% mean includes partial credit. With the GPT-5.4 Mini judge alone, the mean is 98.19% and the binary pass rate 89.47% (119/133). "Record-breaking" is relative to these published results.
- Open source and community: the mobilerun framework is MIT licensed with 9,600+ GitHub stars and 30+ contributors (https://github.com/droidrun/mobilerun), the community Discord has 1,800+ members, the framework reached a #1 day rank on Product Hunt (listed under the Droidrun name), and the homepage shows 2M+ prompts and counting. The hosted cloud (devices, Connect, Agent Mail and the console) is the commercial product on top, with a free plan. Open source: the framework (MIT), Mobile Harness (MIT) and mobilerun Core (Apache-2.0, published on PyPI with its source in the package; no public repository yet).
- Used by teams backed by a16z and Y Combinator (the customers' backers, not mobilerun's). mobilerun's own investors: Merantix Capital, SixtyDegree Capital, TKM and May Ventures (€2.1M pre-seed, July 2025).
- Compliance: not yet SOC 2 or HIPAA certified. SOC 2 is on the roadmap; the healthcare page describes HIPAA and SOC 2 as in progress. Droidrun GmbH is a German company, so EU data protection law applies.
- Company: Droidrun GmbH, Osnabrück, Germany (https://droidrun.ai/). Contact: contact@mobilerun.ai.
- Illustrative figures: values in on-page demos, mock-ups and setup taglines are examples, not measurements or service levels. That covers latencies (p50 204 ms), task times (3.8 s, 14 s, 31 s), "zero to agent in 60 seconds", "real device under five minutes", model names in mock-ups and device or region labels such as US-1 and EU-1.

## Capabilities

- Real Android and iOS devices, recognized as a genuine device
- Real device and agent identity, consistent across sessions rather than regenerated per run: model, serial, IMEI, Android ID, advertising ID and carrier, with GPS, locale and timezone set per device
- mobilerun Connect: residential and mobile network routing configurable by country or region, customers only. Dedicated residential IPs in 21 countries: Australia, Brazil, France, Germany, Hong Kong, Hungary, India, Indonesia, Italy, Japan, Malaysia, Philippines, Singapore, South Africa, South Korea, Spain, Taiwan, Thailand, United Kingdom, United States and Vietnam. Mobile carrier IPs are available on Enterprise, under contract: the pool is smaller than residential, so we scope it with you first.
- eSIM and SMS: bring your own carrier profile, rent a receive-only SMS number (from $60/month, price by country) or request an agent SIM; receive SMS and read conversations over the API on every plan. Sending SMS (outbound) is available on Enterprise
- Agent Mail: claim a permanently reserved inbound email address, list and filter messages, inspect text and attachment metadata, and extract the highest-confidence OTP over the API
- Gmail automation: drive an authorized Gmail account in the real mobile app to read, send, reply and handle attachments; this is separate from Agent Mail
- Persistent state: a stopped phone keeps its apps and data and bills no device minutes; dedicated phones stay on around the clock
- Fleet orchestration (up to 100 concurrent phones self-serve, unlimited on Enterprise), device streaming, screenshots, structured output
- REST API, Python and TypeScript SDKs, a hosted MCP server, and an n8n node

## Plans

Free plan to start. Paid plans from $20/month. Dedicated Android from $64/month. Plans are billed monthly. Dedicated Android is $64/month billed annually ($80 billed monthly) with its own dedicated IP, or $50/month billed monthly without one.

- Free: $0/month. Explore the platform. 60 device minutes (1 h), 500 credits ($5) and 1 GB residential traffic per month. 1 concurrent phone, 1 seat, 1 agent slot, 1 GB storage, Discord support. Usage stops at the allowance, with no extra charges.
- Developer: $20/month. Build and test your automations. 600 device minutes (10 h), 1,500 credits ($15) and 5 GB residential traffic per month. 5 concurrent phones, 3 seats, 10 agent slots, 10 mailboxes, 1 GB storage, support channel.
- Startup: $99/month. Run mobile workflows in production. 3,000 device minutes (50 h), 4,000 credits ($40) and 10 GB residential traffic per month. 25 concurrent phones, 5 seats, 50 agent slots, 25 mailboxes, 25 GB storage, dedicated support.
- Scaleup: $299/month. Scale across teams and workloads. 9,000 device minutes (150 h), 10,000 credits ($100) and 15 GB residential traffic per month. 100 concurrent phones, 20 seats, 200 agent slots, 100 mailboxes, 100 GB storage, dedicated support.
- Enterprise: book a call. Custom limits and SLA, with prices and limits agreed individually. Unlimited concurrent phones (self-serve plans go up to 100) and outbound SMS. Private fleets, custom networking and eSIM provisioning; SOC 2 on the roadmap.

Paid plans continue past the allowance at $0.03 per device minute, $0.01 per credit and $10 per GB, up to a monthly overage limit. You can request a higher limit in billing. Free stops at the limit.

Credits pay for agent usage, like the Assistant and Task Runner. One credit is $0.01. Credits are deducted in real time as agents, chat and the Assistant use models.

## Add-ons

Phone add-ons: dedicated phones and bring-your-own connections, billed on top of your plan, monthly or annually (annual saves 20%).

- Android + dedicated IP: $80/month, or $64/month billed annually ($768/year). A hosted Android phone with its own dedicated residential IP, around the clock. Apps and sessions persist. Dedicated residential IP included. Includes 2,500 credits ($25) per month. Details: https://mobilerun.ai/devices.md
- iPhone + dedicated IP: $200/month, or $160/month billed annually ($1,920/year). A real, physical iPhone reserved for you. Apps and sessions persist. Dedicated residential IP included. Includes 5,000 credits ($50) per month. Details: https://mobilerun.ai/devices.md
- Connect your phone: $5/month, or $4/month billed annually ($48/year). Control your own Android or iPhone from the cloud. Android and iOS. Runs on your hardware. Includes 250 credits ($2.50) per month. Details: https://mobilerun.ai/devices.md

Other add-ons: numbers, IPs, mailboxes and extras, billed on top of your plan.

- Android phone: $50/mo. A hosted Android phone without a dedicated IP, billed monthly. Apps and sessions persist. Add a dedicated IP or bring your own proxy. Details: https://mobilerun.ai/devices.md
- Phone numbers: from $60/mo per number. Receive-only SMS numbers for OTP and verification, delivered to your agent. Price varies by country: United Kingdom $60, United States $70, Hong Kong $80, Australia $100. Sending SMS is available on Enterprise. Details: https://mobilerun.ai/esim.md
- Dedicated residential IP (mobilerun Connect): $20 per 4 weeks per IP. One fixed residential exit in the country you pick, with unlimited bandwidth. Available in 21 countries: Australia, Brazil, France, Germany, Hong Kong, Hungary, India, Indonesia, Italy, Japan, Malaysia, Philippines, Singapore, South Africa, South Korea, Spain, Taiwan, Thailand, United Kingdom, United States and Vietnam. Details: https://mobilerun.ai/connect.md
- Mailboxes (Agent Mail): $2/mo per mailbox. Extra inboxes your agent can read, each with 1,000 inbound messages a month. Developer, Startup and Scaleup include 10, 25 and 100. Details: https://mobilerun.ai/agent-mail.md
- Credits: $0.01 per credit. Prepaid top-ups, used after your plan allowance. Sold in $5 packs (500 credits).

Bring your own proxy on any plan.

Compare all plan features: https://mobilerun.ai/pricing/#compare
Add-ons: https://mobilerun.ai/pricing/#addons

## On the homepage

Give your AI Agent a phone. Plug Claude, Hermes, OpenClaw, or mobilerun VA into a real mobile device to automate any app and receive SMS.

Get started: https://cloud.mobilerun.ai. The hero's Copy Prompt button copies a setup prompt for coding agents that points them at the docs, starting with https://docs.mobilerun.ai/llms.txt. Used by teams backed by a16z and Y Combinator (the customers' backers); mobilerun is backed by Merantix Capital, SixtyDegree Capital, TKM and May Ventures.

### Platform

One platform. The whole mobile stack. Device, agent, identity, network, runtime and observability: one SDK, one API key, one live view.

- **Device: phone fleets.** Up to 100 concurrent phones self-serve, unlimited on Enterprise. Provision Android phones on demand and add dedicated iPhones with an included dedicated residential IP when you need iOS. Add a receive-only number or an inbox when the workflow needs it. Parallel execution, fleet sync, webhooks. https://mobilerun.ai/devices/
- **Agent: describe the task, it runs.** Reads the screen. Taps, swipes and types, and takes a screenshot when it needs to look, so you watch every step live. Drive it from API, SDK, or natural language. https://mobilerun.ai/agents/
- **Identity and runtime: automations that don't stop.** Dedicated, persistent, never shared. Real device, real network. Apps see what they expect, so your agent just keeps going. https://mobilerun.ai/devices/

The model decides. mobilerun executes.

### Why mobilerun

Automation that holds up.

- **24/7 on a dedicated phone**: Add a dedicated Android or iPhone that never sleeps. Your agent runs around the clock, hands-off.
- **AI assistant, no code**: mobilerun VA runs any mobile task in plain language, for the whole team. In beta, paid through credits on any plan.
- **mobilerun Connect**: Residential traffic in every plan, on our own network, never resold. Every layer of a real phone, matched to the device.
- **Human-like typing and swiping**: The agent taps, types and swipes like a human. Fully hosted, fully managed.
- **Build once, replay at scale**: Build an agent once, then replay it as a fast, deterministic script, no model in the loop.
- **Real iPhones**: Real, physical iPhones where simulators can't go.

### Use cases

Four common jobs; the full library is at https://mobilerun.ai/solutions/

- [AI Agent Builders](https://mobilerun.ai/solutions/ai-agent-builders/): A real phone for every agent or every user, with a number and an inbox.
- [Social Media Teams](https://mobilerun.ai/solutions/social-media-automation/): Run authorized brand accounts on real phones.
- [QA Teams](https://mobilerun.ai/solutions/app-testing/): Synthetic users test your app on real phones.
- [Market Intelligence](https://mobilerun.ai/solutions/market-intelligence/): Track competitor apps and AI answers per market.

### Built in the open

The mobilerun framework powers our cloud and runs locally on any device. Open source, MIT licensed, LLM agnostic. Start local, scale to cloud with one config change. `pip install mobilerun` (Python 3.11 to 3.13), then `mobilerun run "Open Settings, check battery"`; add `--vision`, target iOS, or tune steps. Framework: https://github.com/droidrun/mobilerun

- 9,600+ GitHub stars, 1,800+ Discord members, 2M+ prompts and counting, #1 Product Hunt day rank.
- Record-breaking benchmark results, every run public: 98.34% iOSWorld mean official score on 133 tasks, and 100% AndroidWorld official score on 116 tasks (116 successful AndroidWorld trajectories; a successful run on every task, not a single-attempt pass rate). Replays: https://mobilerun.ai/benchmark/
- How it compares: the public iOSWorld leaderboard (https://iosworld.io/, checked September 2026) reports pass rates on the same 133 tasks, with a GPT-5.4 Mini judge and at most 50 interaction steps per task. It is led by Claude Opus 4.6 (vision+XML) at 51.9%, then Claude Sonnet 4.6 at 46.6% and GPT-5.4 at 39.8% (both vision+XML). mobilerun's run is not a leaderboard submission and was set up differently: mobile-harness with app cards and shared memory, a session budget of 30 minutes and corrected data fixtures for 3 tasks. Its binary pass rate is 92.48%: 123 of 133 tasks with a full score of 1.0; the 98.34% mean includes partial credit. With the GPT-5.4 Mini judge alone, the mean is 98.19% and the binary pass rate 89.47% (119/133). "Record-breaking" is relative to these published results.

### Enterprise

Custom infrastructure. Private fleets with no limit on concurrent phones, custom networking, eSIM provisioning, outbound SMS and a custom SLA. iOS and Android in your regions, behind your access controls. SOC 2 on the roadmap. Book a call: https://cal.com/christian-ninstel-bvwhql/15min

---

Source: https://mobilerun.ai/
Full site as markdown: https://mobilerun.ai/llms-full.txt