Bring your customers’ systems into your AI product without building and maintaining the access infrastructure for every integration. Use MarcoPolo for connectivity, task-specific controls, and audit under each customer user’s authority.
Every customer brings different systems, credentials, and access requirements. Your AI feature needs those boundaries enforced as it chooses what to query or do next. Rebuilding connections, inline controls, and audit for each integration takes engineering time away from the feature.
Build the customer experience and agent behavior. Use the SDK and API for access to approved customer systems, with credentials and request enforcement handled in MarcoPolo.
Connects to your customers' systems
Salesforcesales_read
Snowflakeanalytics_ro
Postgresapp_read
Jiraissues_readUse the same application integration pattern with each customer’s approved connections. Keep customer data, source credentials, and end-user grants separate as you onboard additional customers.
Tie access to the customer user and task. Use the activity record to explain which systems the feature reached, what it accessed, and which operations it performed.
Every request attributable to the end user it served
Establish the customer’s connections and end-user authority before the feature requests data. Reuse the integration while configuring access for each customer.
The customer’s administrator or authorized user provisions source connections and approves who may use them. Confirm source permissions and deployment requirements with that customer.
Your application integrates authentication and delegated access so requests run as the correct customer user. Use only connections owned by or granted to that identity.
Your product supplies the feature’s task scope and sends data requests through MarcoPolo. Inline controls further restrict the user’s access, and the activity record stays tied to that authority.
Keep each customer’s connections and user grants separate. A shared application integration does not imply shared customer access.
The task
“Which of our orders shipped late this month, and why?”
The feature answers from the customer’s own data. Other customers’ connections are not granted to this user, and record changes stay outside the task’s scope.
Your product’s agent reads order and shipment records through connections approved for this customer user. The customer’s source credentials stay in MarcoPolo, outside model context.
The application maps the signed-in user to their delegated identity. Only that identity’s approved customer connections are available, and the feature’s task policy limits each request to reads within existing source permissions.
Record what the agent accessed, for which task, and under the customer user’s authority, so your team and the customer’s security team can review the feature’s access.
Architecture, security controls, and deployment documentation to support your security review.
Deploy in your own cloud or on premises, in the region you choose, to meet your data residency, security, and compliance requirements.
Connect your enterprise identity provider for single sign-on. Keep identity and application access managed through your existing IT controls.
Keep source permissions in the systems that own them, including Snowflake and Databricks. Configure additional task restrictions in MarcoPolo; they can narrow access but cannot expand those permissions.
Bring agent activity into your logging and SIEM tools through OpenTelemetry. Trace identity, connection, action, and outcome alongside your existing operational records.
Security brief and compliance documentation on request · Trust & security →
Start with an SDK or API integration for your AI feature. For an OEM partnership, work with us to define deployment, onboarding, operations, and support responsibilities.
