Shared responsibility: platform and customer duties
This page explains how Duale AI and customer teams divide duties across identity, data, model providers, tools, work limits, results, and evidence.
This page covers the shared responsibility split between Duale AI and customer teams, where detailed controls live, external action control, and customer-hosted deployment.
- Responsibility split across identity, data, model providers, tools, work limits, results, and evidence.
- Where detailed controls for credentials, data, models, and evidence are documented.
- Keeping external actions under customer control through SDK Tools and MCP Connections.
- Customer-hosted Platform deployment and its installation agreement terms.
Summaries were generated by AI.
For the managed service, Duale AI operates and patches the Platform infrastructure and runtime. Your team configures Agent Identities and access, writes each Task objective, writes SDK Tool code or selects remote MCP servers, and decides how to use results. A customer-hosted deployment sets its own split through an installation agreement: see Run the Platform in your infrastructure.
Each control has two sides
- Area
- Identity and access
- Duale AI
- Derives the authenticated principal from the credential and enforces applicable Tenant access policies and Library Grants
- Customer team
- Keep credentials secret, review access assignments, define Tenant access-policy requirements, and manage Library Grants
- Area
- Data
- Duale AI
- Applies the documented transport and storage protections to their stated paths and data categories
- Customer team
- Choose what you send and protect your own process, network, and activity cache
- Area
- Model providers
- Duale AI
- Uses only enabled Model Routes in your declared pool and ranks those routes
- Customer team
- Select provider contracts, retention, region, credentials, and the pool
- Area
- Tools and actions
- Duale AI
- Provides Built-in Tools and access to configured SDK Tools and MCP Connections
- Customer team
- Define SDK Tools or choose MCP Connections and their remote principal; authorize protected operations, make repeated execution safe, and add human review when required
- Area
- Work limits
- Duale AI
- Applies documented time and runtime ceilings
- Customer team
- Decide which work you submit, how much runs at once, and total spend
- Area
- Results
- Duale AI
- Validates each answer against the response shape you requested
- Customer team
- Apply domain rules and decide whether the business process accepts the result
- Area
- Evidence and recovery
- Duale AI
- Records part of the Platform activity
- Customer team
- Keep the evidence your process requires, and compensate or repair an unwanted external action
Use the table for planning. It does not vary your contract: the legal documents state the commitments that bind Duale AI.
Where the detailed controls live
- To set who can act: Credentials and lifecycle defines each assurance level and the operations that need it, and token rotation states how a revocation takes effect. Tenant access policies state how direct and group grants combine. Default policies state which supplied tiers Duale AI manages.
- To decide who can reach data, review Data protection and Access and isolation.
- To control model and Tool effects, use Plan the deployment, Tools, and Secure integration.
- To plan review evidence, use Evidence and audit.
Keep external actions under your control
Whatever the deployment shape, you must keep the workload inside the documented scope and limits. Duale AI does not authorize the arguments or operations that can produce External Effects through SDK or MCP Tools. Enforce the applicable deterministic or human decision in your SDK Tool, remote system, or a separate acting process. An MCP Connection does not supply a human-review gate.
Use Agent harness to distinguish autonomy, access, external-action authority, and Effects. Use the threat model when content you do not control, sensitive access, and an external action meet in one Task.
Run the Platform in your infrastructure
Duale AI offers a customer-hosted Platform deployment. Before installation, both parties sign a separate installation agreement that defines the supported deployment:
- release, topology, connectivity, and capacity;
- data path, identity and credential boundaries, and telemetry;
- backup, migration, and upgrade order; and
- support and the responsibility split.
The hosting label alone defines none of these. Write to contact+security@mail.duale.ai to scope the installation. Architecture patterns explains what a separate deployment means for your application.
Return to Security to place this responsibility split in the path of one Task.