File operations

This guide shows you how to read, write, and manage files inside a CoreWeave sandbox instance using the CWSandbox client. It’s intended for developers who build workflows that move data into and out of sandboxes, including parallel uploads and downloads, error handling, and patterns for large files.

Choose authentication

These examples run on serverless capacity unless a section specifies CKS. Choose the credential for the account that owns your sandboxes. See Choose and get credentials to obtain a key or token. Run the installation commands in an active Python virtual environment. Save the Python code from your credential tab as sandbox_auth.py beside the Python scripts you run from this page. It stores the authentication choice, not your credential. The examples import AUTH from that file and pass it explicitly to each new sandbox or session.

W&B API key
CoreWeave API access token

Use your Forge API key through the SDK’s W&B authentication strategy:

uv pip install 'cwsandbox[wandb]>=1.14.2'
export WANDB_API_KEY="[WANDB-API-KEY]"

sandbox_auth.py

from cwsandbox import AuthStrategy

AUTH = AuthStrategy.WANDB

Use the token for your CoreWeave organization:

uv pip install 'cwsandbox>=1.14.2'
export CWSANDBOX_API_KEY="[API-ACCESS-TOKEN]"

sandbox_auth.py

from cwsandbox import AuthStrategy

AUTH = AuthStrategy.COREWEAVE_API_KEY

For CKS placement, choose the CoreWeave token, enable a runner, and set placement_mode="cks" on sandbox creation or in SandboxDefaults. Creating a session with auth=AUTH also authenticates the sandboxes it owns. Keep the same credential when reconnecting, listing, or cleaning up sandboxes.

Basic operations

The examples in this section cover the most common file operations: writing data to a path in the sandbox and reading data back. File operations return OperationRef objects. Use .result() to block for completion.

Write files

from sandbox_auth import AUTH
from cwsandbox import Sandbox

with Sandbox.run(resources={"cpu": "1", "memory": "1Gi"}, auth=AUTH) as sandbox:
    # Write bytes to a file
    sandbox.write_file("/app/data.txt", b"Hello, World!").result()

    # Write JSON
    import json
    config = {"key": "value", "count": 42}
    sandbox.write_file(
        "/app/config.json",
        json.dumps(config).encode()
    ).result()

Read files

# Read file contents as bytes
content = sandbox.read_file("/app/data.txt").result()
print(content.decode())  # "Hello, World!"

# Read JSON
config_bytes = sandbox.read_file("/app/config.json").result()
config = json.loads(config_bytes.decode())

Parallel operations

File operations return immediately, enabling natural parallelism. Running uploads or downloads in parallel can reduce total transfer time when you have multiple independent files.

Parallel uploads

from cwsandbox import results

# Start all uploads simultaneously
write_refs = [
    sandbox.write_file("/app/config.json", config_bytes),
    sandbox.write_file("/app/data.csv", data_bytes),
    sandbox.write_file("/app/model.pkl", model_bytes),
]

# Wait for all to complete
results(write_refs)

Parallel downloads

# Start all downloads simultaneously
read_refs = [
    sandbox.read_file("/app/output.json"),
    sandbox.read_file("/app/metrics.json"),
    sandbox.read_file("/app/logs.txt"),
]

# Get all results
output, metrics, logs = results(read_refs)

Upload-process-download pattern

A common workflow: upload input files, run processing, download results.

from sandbox_auth import AUTH
from cwsandbox import Sandbox, results

with Sandbox.run(resources={"cpu": "1", "memory": "1Gi"}, auth=AUTH) as sandbox:
    # 1. Parallel uploads
    results([
        sandbox.write_file("/app/config.json", config_bytes),
        sandbox.write_file("/app/input.csv", input_bytes),
    ])

    # 2. Sequential processing
    sandbox.exec(["pip", "install", "-r", "requirements.txt"]).result()
    sandbox.exec(["python", "/app/process.py"]).result()

    # 3. Parallel downloads
    output, metrics = results([
        sandbox.read_file("/app/output.json"),
        sandbox.read_file("/app/metrics.json"),
    ])

Error handling

File operations can fail if a path doesn’t exist or isn’t writable. Catch SandboxFileError to handle these cases and inspect the affected path.

File not found

from cwsandbox import SandboxFileError

try:
    content = sandbox.read_file("/nonexistent/file.txt").result()
except SandboxFileError as e:
    print(f"File error: {e.filepath}")

Write errors

try:
    sandbox.write_file("/readonly/path.txt", b"data").result()
except SandboxFileError as e:
    print(f"Cannot write to: {e.filepath}")

Binary files

File operations work with any binary content, including images and serialized Python objects:

# Images
with open("image.png", "rb") as f:
    sandbox.write_file("/app/image.png", f.read()).result()

# Pickle files (only unpickle data from trusted sources)
import pickle
model_bytes = pickle.dumps(my_model)
sandbox.write_file("/app/model.pkl", model_bytes).result()

# Download and unpickle
model_bytes = sandbox.read_file("/app/trained_model.pkl").result()
trained_model = pickle.loads(model_bytes)

Only unpickle data from sources you trust. Loading a pickle file can execute arbitrary code.

Text encoding

Files are transferred as bytes, so you must encode and decode text explicitly to avoid corruption with non-ASCII characters.

# Write text
text = "Hello, Unicode! "
sandbox.write_file("/app/text.txt", text.encode("utf-8")).result()

# Read text
content = sandbox.read_file("/app/text.txt").result()
text = content.decode("utf-8")

Large file considerations

The read_file and write_file APIs are best suited for small to moderately sized files. For large files, consider the following:

Account for bandwidth, because files are transferred through the API.
Use streaming for large datasets when possible.
Access object storage directly instead of transferring the data through the API.

The object_storage_access parameter creates temporary CoreWeave AI Object Storage credentials and provides them to the sandbox’s containers. Your workload then reads and writes buckets directly, so large objects never pass through read_file or write_file. Your organization must first register a Workload Identity Federation configuration with the sandbox control plane; see Object storage access. The following example uses CoreWeave authentication to start a sandbox with read-write access to one bucket, then confirms the credentials were injected. Set CWSANDBOX_API_KEY to a CoreWeave API access token for the configured organization, and replace [BUCKET-NAME] with the name of your bucket:

from cwsandbox import AuthStrategy, ObjectStorageAccess, ObjectStoragePermission, Sandbox

with Sandbox.run(
    auth=AuthStrategy.COREWEAVE_API_KEY,
    container_image="python:3.11",
    object_storage_access=ObjectStorageAccess(
        buckets=["[BUCKET-NAME]"],
        permission=ObjectStoragePermission.READ_WRITE,
    ),
    resources={"cpu": "1", "memory": "1Gi"},
) as sandbox:
    result = sandbox.exec(["printenv", "AWS_ENDPOINT_URL_S3"]).result()
    print(result.stdout)

The sandbox sets the environment variables that AWS SDKs read for container credentials, so any S3-compatible client inside it works without explicit keys. For those variables and a Boto3 example, see Use the credentials inside the sandbox. To limit the credential to a key prefix, use object_prefix. For workloads that only read data, use ObjectStoragePermission.READ. For large temporary data that stays local to the sandbox, attach a volume instead. See Volumes.

Sandboxes

Use sandboxes

Manage sandboxes

Reference

Choose authentication

Basic operations

Write files

Read files

Parallel operations

Parallel uploads

Parallel downloads

Upload-process-download pattern

Error handling

File not found

Write errors

Binary files

Text encoding

Large file considerations

​Choose authentication

​Basic operations

​Write files

​Read files

​Parallel operations

​Parallel uploads

​Parallel downloads

​Upload-process-download pattern

​Error handling

​File not found

​Write errors

​Binary files

​Text encoding

​Large file considerations

Choose authentication

Basic operations

Write files

Read files

Parallel operations

Parallel uploads

Parallel downloads

Upload-process-download pattern

Error handling

File not found

Write errors

Binary files

Text encoding

Large file considerations