Storage and secrets

A file browser for every volume, and API keys that reach containers as environment variables

Storage

Storage lists the workspace’s volumes with their size, file and folder counts, and where each one is mounted (every function, box or notebook whose code references it). Create a volume from the Create volume dialog or with levers9 volume create NAME.

Open a volume for a file browser:

• folders, with breadcrumbs;
• upload (a file or a folder) and download;
• previews for text, code, images, audio, video and PDF. A file is served for what it is and never runs as a page of the dashboard: pictures, sound, video, PDF and plain text open in the browser, and anything else (HTML, SVG, XML, JSON, scripts) is downloaded when opened in its own tab, while the preview still shows it as text or as a picture;
• delete from the row menu (levers9 mv renames or moves inside a volume);
• a reference list showing which apps mount the volume and at what path.

The same volume is what your code sees at mount_path:

from levers9 import Volume, job

@job(volumes=[Volume(name="datasets", mount_path="/data")])
def count():
    import os
    return len(os.listdir("/data"))

levers9 ls, mv and rm work on VOLUME/path from a terminal, with shell wildcards; levers9 cp ./file levers9://VOLUME/path uploads and the reverse downloads (see the CLI reference). A CloudBucket mounts an S3 or GCS bucket the same way, without copying.

Notebooks keep their files on the notebooks volume under /notebooks/<name>; datasets live on the datasets volume as TSV files; RL Arena preference snapshots land there too.

Secrets

Secrets stores API keys and other values for the workspace. Create secret offers templates for common integrations (Anthropic, OpenAI, Hugging Face, AWS, Google Cloud, Google Sheets, GitHub, Slack, or custom variables), each pre-filled with the variable names the integration expects and a link to where to find them. Rotate a value in place; containers pick up the new value on their next start. Delete when it is no longer needed; functions that read it fail on their next start.

A container sees a secret as an environment variable when it asks for it:

@job(secrets=["OPENAI_API_KEY"])
def ask():
    import os
    key = os.environ["OPENAI_API_KEY"]

Datasets bind provider credentials through the same secrets, so one client’s key never reaches another client’s build. Notebooks attach secrets the same way from their profile panel.