> ## Documentation Index
> Fetch the complete documentation index at: https://docs.anyreach.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Code step
> Run sandboxed Python inside a workflow.
The Code step runs Python in a sandboxed environment. Use it when no other step type fits — complex data transformations, custom logic, calling Python client libraries, or looping over arrays.
## When to use it
| Use Code when | Use something else when |
| ------------------------------------------------------------------- | ----------------------------------------------------- |
| Transforming data shapes (reshaping JSON, computing derived fields) | A single HTTP call is enough — use HTTP API step |
| Looping over an array | A simple branch is enough — use Condition step |
| Calling a Python client library (`boto3`, `stripe`, etc.) | The integration exists in Pipedream — use Action step |
| Logic too complex for a JSONata expression | The transform is one line — keep it inline |
## How it works
Your code runs inside a function called `run(ctx)`. The current workflow context is injected as `ctx` — a Python dict containing every upstream step's output. Whatever `run` returns becomes this step's output in `ctx`.
```python theme={null}
def run(ctx):
contact = ctx["Lookup contact"]["body"]
return {
"display_name": f"{contact['first_name']} {contact['last_name']}",
"is_premium": contact.get("tier") in ("gold", "platinum"),
}
```
## Inspector
The Code step inspector has two tabs:
* **Configuration** — Draft your Python code directly in the provided editor. For an enhanced experience, click the purple **Open IDE** button to access a full-screen coding environment.
* Here, you can write and edit your Python logic with greater visibility.
* Specify any external Python packages your code requires—these will be installed into the sandboxed environment automatically.
* Here you can write your pythion code
* Enter packages that needs to be installed in the sandbox
* **Output** — After saving your code in the editor, this tab displays the structure of the data returned by your `run` function. You can reference these output fields as variables in downstream steps using the workflow context.
## Accessing context
The `ctx` dict contains all upstream step outputs keyed by step name:
```python theme={null}
def run(ctx):
# Access trigger/input data
email = ctx["input"]["customer_email"]
# Access upstream step output
contact = ctx["Lookup contact"]["body"]
status = ctx["Lookup contact"]["status_code"]
return {"email": email, "name": contact["first_name"]}
```
## Examples
### Reshape an HTTP response
```python theme={null}
def run(ctx):
contact = ctx["Lookup contact"]["body"]
return {
"display_name": f"{contact['first_name']} {contact['last_name']}",
"tags": [t["name"] for t in contact.get("tags", [])],
"is_premium": contact.get("tier") in ("gold", "platinum"),
}
```
### Call a third-party SDK
```python theme={null}
import stripe
def run(ctx):
stripe.api_key = ctx["input"]["stripe_key"]
customer = stripe.Customer.retrieve(ctx["input"]["customer_id"])
return {
"email": customer.email,
"balance": customer.balance,
}
```
(Add `stripe` to the packages list in the inspector.)
## Limits
| Limit | Value |
| --------------------- | --------------------------------------------------- |
| Execution timeout | 300 seconds |
| Max return value size | 10 MB |
| Network | Outbound HTTPS allowed |
| Filesystem | Writable scratch space — not persisted between runs |
## Considerations
* **Don't cache state between runs.** Each run gets a fresh sandbox.
* **Let exceptions propagate.** Caught-and-swallowed exceptions hide bugs. If you handle an error, return a structured `{"failed": True, "reason": "..."}` and branch on it with a Condition step.
* **Avoid heavy packages on the critical path.** `pandas` and `numpy` add cold-start latency. For simple data shaping, write it without dependencies.