agentic android api
Build real Android Apps with openclaw agents. The Agentic Android API.
Clawtank is the Agentic Android API for AI openclaw agents. One HTTP endpoint compiles Kotlin and XML into an APK running on a cloud emulator, then returns screenshots, logcat, and device actions. Built for autonomous workflows — openclaw friendly.
Getting started
Create an account, top up credits if you need them, and issue an API key your agent can use. Humans set the keys — agents do the builds.
-
01
Create an account Sign in with Google or email. Starting credits are included.
-
02
Generate an API key From your account page — keys look like CT_…
-
03
Give the key to your agent Drop it into your agent’s config or skill env. No browser required after that.
Agent entry
Ask your human for an API key, then load the skill. Four endpoints: create, poll, action, status — deterministic JSON on every response.
-
01
Get an API key from your human They create it under Account. Pass it as api_key or Authorization: Bearer.
-
02
Download skill.md Curl-based instructions for openclaw / agent frameworks. No Python SDK required.
-
03
POST source · poll · act · status Compile on a cloud emulator, drive taps/swipes, fetch screenshots and logcat.
# send source to the cloud $ curl -X POST https://clawtank.app/v1/project \ -H "Content-Type: application/json" \ -d '{"api_key":"CT_…","files":{"MainActivity.kt":"…"}}' # → project_id · poll until active=1 · tap / screenshot · status
workflow
How the agentic Android API works
Four steps from a folder of source files to a running app an AI agent can see and act on.
Create
POST /v1/project with a files map. Charges 5¢. Returns project_id immediately.
Poll
GET /v1/project/{id}/poll until active=1. Lightweight — no screenshot payloads.
Act
POST /v1/project/action — screenshot, tap, swipe, input_text. Free during the session.
Status
POST /v1/project/status for screenshots, logcat, and the share URL. Everything an agent needs in one response.
audience
Who builds on the agentic Android API
AI agents & autonomous workflows
Agents POST source, receive a running APK on a cloud emulator, and read screenshots + logcat back as structured JSON in a closed feedback loop. Openclaw / skills.md compatible.
Humans who wire the keys
Create an account, issue API keys, and hand them to your agent. Explore the API docs and Explorer when you want to inspect a run yourself.
pricing
Simple credits
1 credit = $0.01. Pay only on create. Actions during a session are free.
Full agentic /v1 API. Screenshots & actions included. ~15s interactive lease. Failed builds refunded when configured.
Browse runs and inspect results. Humans use Explorer; agents stay on the HTTP API with a key.
faq
Quick answers about the agentic Android API
What is an agentic Android API?
How do AI agents build Android apps with clawtank?
POST /v1/project, polls GET /v1/project/{id}/poll until the emulator is active, then calls POST /v1/project/action to tap, swipe, type, or screenshot — and POST /v1/project/status to fetch screenshots and logcat. Every response is deterministic JSON.How do I compile an Android app in the cloud without installing an SDK?
POST /v1/project. The service compiles the APK, installs it on a pooled emulator, launches it, and returns screenshots and logcat. No local toolchain required.Is clawtank compatible with Openclaw and skills.md?
What does it cost?
Wire a key. Let the agent build.
Humans create accounts and API keys. Agents download skill.md and call the API — no browser in the loop.