Contract Autopilot
Learn more about Contract Autopilot, a Soda AI-powered feature to create data contracts tailored to your data.
Contract Autopilot is currently available as a preview feature. Contact support@soda.io for early access.
Contract Autopilot automatically generates initial data contracts for datasets with recommended data quality checks to accelerate onboarding in Soda Cloud. It is designed to help teams achieve fast time to value and iterate later.
New: run Autopilot with Soda AI. Point Soda AI at your datasets to draft fully populated contracts with recommended checks from your data's own profile. You approve what gets enforced.
Read more about Soda AI's capabilities.
Contract Autopilot automatically generates initial data contracts for datasets with recommended data quality checks to accelerate onboarding in Soda Cloud. It is designed to help teams achieve fast time to value and iterate later.

Watch Contract Autopilot in action
💡 Contract Autopilot vs. Create Contract
Contract Autopilot analyzes sampled data and profiling metadata to automatically generate a fully populated contract with recommended checks.
When you click Create Contract (or use the CLI generate command), Soda creates a contract skeleton based on the dataset schema. It includes columns only and no checks are added.
Generate a contract with Autopilot
This feature requires the Manage contract permission on the select datasets. Learn more about permissions here: Dataset Attributes & Responsibilities
You can generate a contract with natural language while chatting with Soda AI:
You can now view the generated contracts to review, edit or suggest changes.
You can generate a contract manually by selecting the datasets you want to monitor:
[Optional] Set a schedule for contracts
You can optionally define a schedule to run contract verification at a time of your choosing.
Use the data source's default schedule if your data source has a default schedule enabled.
Manual schedule if your data source does not have a default schedule, or if you wish to manually set a schedule for the generated contracts.

You can now view the generated contracts to review, edit or suggest changes.
This process will create a scan for each selected dataset. You can cancel the scan or review the execution logs in the Scans tab on the dataset, as well as on the general Scans page.
Successful scan
The generated contract is published.
A purple checkmark icon indicates that it is an automatically generated contract.
The icon will change to a blue checkmark icon after there is a user edit, to indicate it is a user-generated contract.
A data contract history entry is published, indicating that Autopilot published a first version of the contract.
[If configured] A "published contract" Webhook will be published (for example, to allow sync to Git).
Failed scan
The dataset is marked with the message "Contract generation has failed".
Scan logs can be reviewed to understand the underlying cause.
The contract generation can be re-triggered after inspecting the root cause.
Requirements & limitations
Contract Generation Scope
Autopilot can only generate data contracts for datasets that do not already have an existing data contract. Datasets with an existing contract are not eligible for contract generation via Autopilot.
Autopilot is currently a Soda Cloud feature:
A Soda Runner is required
CLI or Python API are not available
APIs are not available
Data types
Autopilot works on the following data types:
text
numeric
timestamps
Columns with other data types, such as JSON or structured array, for example, are not currently supported.
Data sources
Autopilot is supported on:
Snowflake
Databricks
Postgres
Athena
Trino
Dremio
Oracle
SQLServer
Synapse
Fabric
Redshift
Contract Autopilot is assistive and does not autonomously modify data or execute destructive operations. Users can review generated contracts before execution.
Privacy & security
Contract Autopilot is designed to minimize data exposure while generating meaningful contract recommendations.
The following data is used locally by Soda:
A sample of approximately 10,000 rows to compute profiling metrics
The following data is sent to Soda’s managed OpenAI API:
Computed profiling metrics
Data source type
Dataset schema
A sample of up to 100 rows, if enabled in the organization settings.
Learn more about Soda's AI security & privacy measures.
Next steps
Manage the organization settings to:
enable/disable Contract Autopilot
allow Soda to send a small sample of row data to OpenAI
After contract generation, you can:
Review generated checks
Validate assumptions with domain experts
Adjust schedules and thresholds
Iterate using Contract Copilot
Last updated
Was this helpful?






