csvpile Help

API

The public API is read-only and returns dataset files by numeric dataset ID.

• GET /api/datasets/:datasetId
• GET /api/datasets/:datasetId/versions/:versionNumber
• Formats: csv, tsv, jsonl

The dataset ID stays the same when you publish new versions or restore an older version.

curl -L \
  -H "x-api-key: YOUR_API_KEY" \
  "https://acme.csvpile.com/api/datasets/42?format=jsonl" \
  -o dataset-42.jsonl

• 401: missing, invalid, or revoked API key.
• 403: wrong auth type, such as a user session instead of an API key.
• 404: dataset not found, or not in the API key’s workspace.

MCP

Read and update datasets from supported AI chat clients. viewer users get read-only access.

• Server URL: https://mcp.csvpile.com/api/mcp
• Auth: OAuth 2.1 with PKCE

1. Open MCP settings in your chat client.
2. Paste your workspace server URL.
3. Sign in to csvpile upon redirect to approve access.

Usage examples:

• Build an Eval Dataset from Scratch "I'm building a customer support bot that handles refunds, order tracking, and account changes. Create a dataset that covers the key scenarios I should be evaluating, including tricky edge cases."
• Audit and Fill Gaps in an Existing Dataset "Check the coverage of my billing-agent-evals dataset. Where are the blind spots? Add rows to cover anything we're missing."
• Iterate When Production Breaks "Here are 5 prompts from production where our agent hallucinated a return policy. Add these to my returns-evals dataset and generate 10 similar variations so we catch this pattern going forward."

Import from Hugging Face

Some imports may fail if the source is unsupported or too large.

1. Find the dataset on Hugging Face and copy the dataset name (e.g. openai/gsm8k).
2. Paste the dataset name into the import field.
3. Choose a config or split if the dataset has more than one.

Dataset Management

Use csvpile to organize datasets, update them safely, and keep version history in one place.

• Create folders to group datasets and move datasets between folders.
• Rename datasets without changing their stable API ID.
• Replace a dataset file to publish a new latest version.
• Restore an older version when you need to roll back.
• Use checkout and check-in locks when editing to avoid conflicting changes.

Replacing or restoring a dataset keeps the same dataset ID for API access. Very large datasets may be preview-only in the web editor.

Self-Hosted Storage

By default csvpile manages storage for you. If you want to use your own S3 bucket, you can.

1. In AWS, create a new bucket and create an IAM user with read/write access to the bucket.
2. Bucket name: enter the actual bucket name you created in AWS. Not the bucket ARN.
3. Folder path: optionally enter the folder path you want to use if not root, e.g. datasets
4. Region: enter the AWS region you created the bucket in. e.g. us-east-1
5. Access key ID: enter the access key ID you created for the IAM user.
6. Secret access key: enter the secret access key you created for the IAM user.
7. Remember to click "Save and use Self-Hosted" to apply the changes.
8. Switching storage affects future uploads only. Existing versions stay in the storage location where they were written.

Team Access

Workspace membership is managed from the organization members screen.

• admin: full workspace access, including API key and storage settings.
• editor: can create, edit, replace, and restore datasets.
• viewer: read-only in the web app and read-only in MCP.

Enterprise SSO

Enterprise SSO is available for eligible workspaces. csvpile sets this up during onboarding.

Contact Us

If your question is not answered above, please email help@csvpile.com. csvpile was built by Bouldly Typed, LLC, a software development and consulting company.