Box Mount

Use Box Mount as mounted storage for your agent workflows. It exposes a Box folder as a local filesystem inside a Runta runtime, so agents can read source files and write results with normal file operations. Box Mount handles synchronization with Box; Runta provides the execution layer.

The sales report example uses the Runta Python SDK to create a runtime and run an OpenAI Agent. The Agent reads store sales CSVs from the mounted folder, calculates totals with Decimal, and writes an Excel workbook and a daily report back to Box. It can also assign the daily report to a human reviewer using Box Tasks.

Set Up

You need uv, Python 3.12 or later, Linux or macOS for the local orchestrator, and a Linux x86_64 Box Mount archive matching the example’s runtime architecture.

git clone https://github.com/runta-dev/box-integration-example.git
cd box-integration-example
uv sync
cp -n .env.example .env

Fill in .env:

Variable	Value
RUNTA_TOKEN	Your Runta API token.
BOX_ACCESS_TOKEN	A Box token with read/write access to a dedicated test folder.
BOX_FOLDER_ID	The parent test folder ID.
BOX_MOUNT_ARCHIVE	Local path to the Linux x86_64 Box Mount archive.
OPENAI_API_KEY	Your OpenAI API key.
OPENAI_MODEL	A model ID with tool-call support, available to your account.
BOX_REVIEWER_USER_ID	Optional numeric Box user ID for human review. The user must already have access to the report.

Keep credentials in your local .env file.

Run the Demo

uv run doctor  # Check prerequisites and service access.
uv run seed    # Prepare sample CSVs in the Sales Report Demo folder.
uv run demo    # Mount Box, generate reports, optionally assign review, and clean up.

Each command shows checkbox progress. The workflow is:

1. Mount the Box folder inside the Runta runtime.
2. Let the Agent read Incoming/, validate CSV rows, calculate totals, and write reports.
3. If a reviewer is configured, find the generated daily report in Box and create and assign a review task.
4. Normally unmount the folder and delete the runtime.

Reports remain in Box under Sales Report Demo/Results/<date>/<run-id>/: sales-report.xlsx, daily-report.md, summary.json, and agent-run.json.

Failed runs retain the runtime ID and logs. Use the state directory printed by the command to retry review assignment and clean up:

uv run python demo.py --state-dir /path/to/state cleanup