Anywhere untrusted or AI-generated code needs to run, Cinch is the sealed, disposable layer that executes it and returns just the result. Here’s where teams reach for it.
The core use case
AI agents & LLM tools
Your model writes code. Cinch runs it.
When an LLM generates Python or JavaScript, you cannot trust it enough to run on your own servers — one bad snippet can read your secrets, hit your network, or crash your host. Cinch gives your agent a sealed, disposable box to execute in, and returns just the output. Wire it into a tool call and your agent can compute, transform data, and test its own code safely.
Code-interpreter tools · autonomous agents · AI data analysts
Tutorials, documentation, and courses land harder when readers can run the examples live. Drop a Cinch call behind a “Run” button and execute user-submitted snippets without spinning up infrastructure or worrying about what they type. Every run is isolated and throwaway.
Docs with runnable examples · coding courses · REPLs
Running homework you did not write is exactly the untrusted-code problem Cinch solves. Execute each submission against your test cases in a fresh sandbox, capture the output, and grade it — without a fork bomb or infinite loop ever touching your graders.
Run customer-supplied transforms without touching your infra.
Let users bring their own scripts to transform, filter, or compute over data inside your app — without you hosting arbitrary code. Cinch runs the logic in isolation and hands back the result, so a customer’s code can never reach your systems or other tenants’ data.
Coding assessments need to actually run candidate code — safely, and for many people at once. Cinch executes each submission in its own sealed box with per-run limits and a hard concurrency ceiling, so a heavy assessment window never turns into a security or capacity problem.
Give your users an escape hatch: a custom-code step that runs real Python or JavaScript inside your workflow builder or app — without you executing untrusted logic on your own servers. Cinch is the safe execution layer underneath the “add a code block” button.
Every use case above is the same call under the hood: send code, get back the output — isolated, metered, and disposable. You bring the product; Cinch handles the dangerous part.
from pangolin import Sandbox
box = Sandbox(api_key="cinch_live_...")
result = box.run("print(2 + 2)")
print(result.stdout) # "4"
Whatever you’re building, run the code safely.
Grab an API key and run your first sandbox in minutes. Free to start.
