Skip to content

Resource · Governance

Agents can move quickly without becoming invisible administrators.

ConfigAI governance is built around evidence, exact revisions, explicit review gates, durable task history, and protected product data. Human authority is concentrated at the moments that change what customers can configure or what product data becomes active.

01

Evidence

Product facts retain source links, revisions, and captured evidence instead of becoming unattributed model output.

02

Review gates

Product Apply, 3D attach, semantic migrations, and rule publication are explicit decisions.

03

History

Previous product, field, model, and rule revisions remain inspectable rather than overwritten.

Authority model

Drafting authority and publishing authority are different permissions.

Agents and deterministic importers need freedom to produce useful proposals. They do not need permission to silently publish those proposals into active customer configuration data.

Draft

Agent / importer

Research, capture, normalize, draft, test, explain, and submit a proposal.

Review

Reviewer

Inspect source evidence, identity matches, semantic impact, assumptions, test results, and diffs.

Controlled

Apply transaction

Commit the approved product/model/rule revision under an idempotent mutation identity.

Read

Runtime consumer

Configurators and agents consume only the reviewed revision permitted for that workspace.

Protected data

Test cleanup is never allowed to guess which database is disposable.

Runtime and recovery databases can contain real application data even when their names sound temporary. Cleanup must target only the exact database created by that test invocation.

01Identify exact runtimeResolve the active ConfigAI profile, service ownership, and database identity before destructive operations.
02Protect canonical & recovery dataConfigured runtime/recovery databases require explicit authorization and verified backup before removal.Fail closed
03Isolate testsA test may clean up only the database it successfully created and recorded for that invocation.
04Treat missing data as recoveryDo not silently replace a missing product database with an empty schema or older copy.

Credential boundary

User account connections should follow the user without leaking into CPQ workers.

Account-level OAuth and agent-runtime credentials belong to the signed-in user/platform boundary. Workspace bindings and product-domain workers receive only the scoped capabilities they require.

User account connector

OAuth/credential lifecycle belongs to the signed-in account rather than a CPQ product record.

01

Workspace runtime binding

Each workspace chooses which runtime/agent may execute its tasks.

02

Scoped agent task

The task receives workspace/product scope and approved tools, not unrestricted customer credentials.

03

CPQ service

Product state remains independent from provider credential storage.

04

ConfigAI

Fast automation with visible authority.

ConfigAI can let agents do more work because the source, revision, task, and human decision boundaries stay explicit.

Explore agent setup