Guides: every agent workload,
end to end.
Complete walkthroughs — not snippets. Each guide takes you from zero to a working setup with the CLI, the Python/TS SDK, or raw HTTP. Try any request live in the interactive API reference. New accounts start with $5 free credits.
Guide 01 · template: code-interpreter
Build an AI code interpreter for data analysis
Give your LLM a Python runtime it can actually execute in: upload a CSV, let the model write pandas, run it safely, and pull the chart back out. This is the classic "code interpreter API" loop behind ChatGPT-style data analysis.
Guide 02 · templates: claude, claude-terminal, codex, opencode, amp, grok
Run Claude Code & autonomous coding agents safely
A coding agent wants a full shell, a repo checkout, and hours of uninterrupted time — exactly what you don't want on your laptop or prod boxes. PillDrops ships ready templates for the five major agent CLIs; the agent gets root in its own microVM, you get a wall.
Guide 03 · templates: playwright, playwright-headed
Browser automation & web scraping for AI agents
Agents that browse need Chromium, and Chromium needs a machine. The
playwright template boots with the browser and drivers installed — write the
script (or let the model write it), run it, download what it found.
Guide 04 · template: desktop
Computer-use agents on a real Linux desktop
Computer-use models (Claude computer use, OpenAI CUA) click and type on real
software. The desktop template boots a full Linux desktop streamed over VNC —
screenshot it, feed the frame to the model, execute its clicks.
Guide 05 · pause / resume / refresh
Long-running background agents that don't burn money
Research agents, monitoring agents, multi-hour builds: sessions run up to
24 hours on paid plans, extend live with refresh, and — the part that saves real
money — pause when idle so a waiting agent costs storage, not compute.
Guide 06 · 1,000 concurrent, stress-tested
Run evals and RL rollouts at scale
Benchmark 500 model outputs, roll out RL episodes, grade a class of solutions — each in its own clean environment. The multi-node scheduler bin-packs sandboxes across the fleet; the history API gives you the exact cost of every single run.
Guide 07 · deploy in one call
Instant full-stack previews of AI-generated apps
Building a v0-style app generator? One API call takes your agent's output from
files to a running server with a shareable HTTPS URL — files in, deps installed, process
started, app_url back.
Guide 08 · $0.0000002 per invocation
Turn agent tools into HTTP functions
A tool your agents call thousands of times shouldn't boot a sandbox each time. Register the handler once as a function; PillDrops keeps it warm behind an HTTP endpoint and meters per call.
Guide 09 · multi-tenant isolation
Run untrusted user code in your product
Plugin systems, coding-education platforms, user-defined webhooks — anywhere strangers' code runs, it must not run on your infrastructure. Give each execution a microVM with hard caps and a deadline.
Guide 10 · exit codes preserved
CI for AI-written code — test before you trust
Your agent opened a PR at 3 a.m. Run its code through the tests in a disposable
sandbox from CI — pilldrops exec exits with the remote command's exit code, so
pipelines just work.
Guide 11 · disposable compute
Data jobs & ETL without a job cluster
Nightly transforms, one-off backfills, agent-triggered data pulls: boot a sandbox sized for the job, run it, download the output, destroy it. No queue, no cluster, no idle workers.
Guide 12 · any Dockerfile
Ship your own toolchain with custom images
When the 16 templates aren't enough — private deps, system packages, pinned toolchains — build a custom image from any Dockerfile and launch sandboxes from it like any template.
Guide 13 · templates: mobile-test / android / android-headless · requires a KVM node (Pro plan+)
Mobile E2E testing on a cloud Android emulator
A device farm without the farm: each sandbox boots a KVM-accelerated
Android 14 emulator with adb and Appium inside. Run your suite with one
command and get a session page — live progress while it runs, then the full report
with a screen recording — or drive the device by hand.
Guide 14 · template: claude · no API key needed
Use your Claude Pro/Max subscription — no API key
Already paying for Claude? One command in the workstation's web terminal
(claude setup-token) signs the machine in with your subscription — no API key,
no per-token bill, nothing for us to store.
Guide 15 · template: codex · no API key needed
Use your ChatGPT Plus/Pro subscription with Codex
Codex signs in with the ChatGPT account you already have:
codex login --device-auth, open the URL, type the one-time code. The CLI itself
recommends this on remote machines — your cloud workstation is exactly that.