Read: Start with one business decision
For Buyers · Nodes

Define authority before data access

·Sources checked Sep 21, 2026

Before access begins, agree what each person and assigned agent may see, reveal, recommend and change, plus the approval and exception owners. Nodes supports Nodes Cloud, a single-tenant customer VPC and customer-managed on-premises deployment. Review data, model and operational paths for the chosen boundary. Nodes works within configured permissions. Actions requiring approval remain gated by the customer's policy.

Compare the three operating boundaries

Nodes Cloud runs inside a Nodes-managed boundary governed by the customer agreement. A customer VPC deployment runs single-tenant inside an approved AWS, Azure, or GCP environment. An on-premises deployment runs on customer-managed infrastructure. Deployment flexibility is a platform capability, so the security review must attach each claim to the option under consideration.

For a private VPC or on-premises deployment, the approved boundary can keep customer production data, customer-specific intelligence, Decision Traces, and deployed model artifacts inside the customer environment with zero customer production-data egress. The review should still document any approved model endpoint, telemetry path, support access, artifact release, backup, and recovery mechanism instead of inferring it from the deployment label.

The customer owns the separable knowledge graph, context and history defined in the agreement, including evidence, relationships, human judgments, decisions and outcomes. Customer-specific weights apply where present. Confidential memory is not pooled across tenants by default. Any permitted model or artifact release requires an explicit boundary and approval process. Ownership and exit rights are reviewable before production.

Integration has a stop condition

Inspect proposed mappings, their uncertainty and human verification. Confirm integration feasibility, permitted reads and writes, dependencies and system owners for the scoped workflow. Demonstrate that a renamed or drifted field pauses affected work and reaches the exception owner. Agree implementation and maintenance responsibilities for that environment.

That mechanism gives the platform team four concrete review points:

  • agreed sources, company-selected checking frequency, permissions and data paths
  • the proposed field mapping and its confidence
  • the human acceptance record for uncertain mappings
  • the behavior when a schema changes or a source becomes unavailable

Every recommendation carries its evidence

A Decision Trace links evidence, recorded reasoning, versions, human corrections, authorized action and later outcome. Inspect the records for the scoped application within access rights. Named people control consequential decisions and production changes. Routine execution remains within current permissions; a new objective or historical approval cannot expand them. Designated workflows can require a second signer.

Map each proposed action to its tools, current permissions, required approval, exception owner and stop conditions. Test that a declined action stays stopped. Agree quality evidence, exception limits and total AI and retained human cost, including correction and adoption. The owner decides whether to expand, hold or narrow delegation after those checks.

Compare supported model options against the same task-quality, exception, cost and permitted-data requirements. Model changes need evaluation and configuration; retained company knowledge is distinct from the model. The decision record supports security and audit review without establishing legal compliance on its own.

One deployment, with clear limits

At one Fortune 500 insurance carrier, six AI hiring vendors had been rejected over 18 months on architecture. Nodes completed legal approval in 17 days.

The same deployment moved from contract to production in 34 days. These are results from one deployment, live since January 2025. They are evidence of the mechanism, not a timeline promise for another environment.

Frequently asked questions

Why buy this if the data team has already joined the systems?

Joined tables connect data; a company context graph also connects business relationships, human judgment, decisions and outcome history. Nodes Engine builds and uses this graph to investigate, compare options and carry out approved work. You buy Engine, Connector and AI FDE while retaining company rules, graph and history. Compare integration, review surfaces, recovery and outcome evaluation on the same scope, including work your team retains. Compare internal build and enterprise platforms.

Does fast integration mean controls were skipped?

Ask to inspect the mechanism and acceptance tests. Proposed mapping, human verification, scoped permissions, and a stop on schema drift must be demonstrated in your environment. The 17-day and 34-day results remain one-deployment evidence, not proof of automatic implementation for a new workflow.

What happens if the relationship ends?

The customer retains the separable intelligence and applicable model artifacts defined in the agreement. Request a sample export with stable identifiers, relationships, schemas, and provenance; test what remains usable without Nodes. Agree formats, retention, and transition support. Nodes' platform code and licensed runtime remain separate, and operational services end according to the contract.

Review the proposed deployment and controls

Bring one question or workflow, its owner and deployment requirements. Review access and recommendation scope before analysis, then separately agree any implementation authority. Confirm data and model paths, stop conditions, retention and relevant diligence materials. No company data is required for the first conversation. Request a security review or inspect the security materials and their availability.