Make a python file that uses the chalkcompute package (PyPI name is chalkcompute, not chalk-sandbox-sdk). Use a uv shebang. Pin Python to <3.14 because a transitive Rust dep (chalk-remote-call-python) builds against PyO3 which currently maxes out at 3.13.
#!/usr/bin/env -S uv run --script
#
# /// script
# requires-python = ">=3.12,<3.14"
# dependencies = ["chalkcompute"]
# ///
import chalkcompute
@chalkcompute.function()
def some_function(arg_1: str) -> str:
return f"hi {arg_1}"
if name == "main": print(some_function("world"))
Notes:
- `@chalkcompute.function` must be **called** — `@chalkcompute.function()` (keyword-only args). Bare `@chalkcompute.function` raises `TypeError: function() takes 0 positional arguments but 1 was given`.
- The decorator deploys the function at import time (builds image, uploads, registers).
- Calling the decorated function (`some_function("world")`) invokes it remotely and returns the result.
## Required env vars
The SDK reads creds from env, **not** from `chalk` CLI config. The env var is `CHALK_ENVIRONMENT_ID`, not `CHALK_ENVIRONMENT`.
CHALK_API_SERVER CHALK_CLIENT_ID CHALK_CLIENT_SECRET CHALK_ENVIRONMENT_ID
You can read these out of `chalk config` output for the current project. Example invocation:
CHALK_API_SERVER=... CHALK_CLIENT_ID=... CHALK_CLIENT_SECRET=... CHALK_ENVIRONMENT_ID=... ./your_script.py
Without `CHALK_ENVIRONMENT_ID` you'll get a `permission_denied` error from `CustomImageService.GetOrBuildCustomImage` even though `CHALK_CLIENT_ID`/`CHALK_CLIENT_SECRET` are valid.
## Local SDK source
If `chalkcompute` isn't on the configured index (e.g. the script can't resolve `chalkcompute` from PyPI), point uv at a local checkout of `chalk-sandbox-sdk`: