The control planeto govern
Move AI systems from experiment to production with named owners, explicit policy, reversible releases, and an audit trail that stays attached to every change.
Control what moves.
Understand why it moved.
Policy-bound releases
Attach model, data, tool, and environment policy to a release before it can move beyond staging.
Evaluation evidence
Keep test suites, reviewer decisions, and known limitations beside the exact system version they describe.
Human release gates
Route sensitive changes to the right owner, record the decision, and make exceptions visible instead of implicit.
Operational boundaries
Scope tools, credentials, data access, and runtime permissions independently for each AI system and environment.
Three deliberate steps.
No invisible hand-offs.
1import { arcwise } from '@arcwise/control'23arcwise.define({4 system: 'support-router',5 tools: ['knowledge.read'],6 escalation: 'human-review'7})
Boundaries by
default.
Arcwise treats every environment as a distinct contract. Data, tools, approvals, and rollback rules travel with the release instead of living in a separate checklist.
Governance you
can inspect.
Put existing systems
inside the policy boundary.
Reference connectors shown for interface planning. Availability depends on implementation.
Control is
designed in.
Arcwise is presented here as a product concept—not a certified or operational service. The interface shows how release evidence and responsibility could stay legible.
Policy checks
Evaluate the declared data, tool, and runtime boundary before a version can advance.
Least-privilege tools
Scope each system to the narrowest set of actions required for its assigned workflow.
Human checkpoints
Require a named decision when a release crosses a sensitive operational boundary.
Decision context
Retain evaluation results, exceptions, release notes, and rollback intent together.
Policy as code.
Review as a team.
A compact interface study for defining boundaries, attaching evaluation evidence, and requesting a release. The package and examples are illustrative only.
Typed policy
Model permissions as reviewable code.
Explicit defaults
Make inherited behavior visible.
Provider-aware
Keep model and runtime context together.
Evidence-linked
Trace decisions to the evaluated version.
// Illustrative API — not a live packageimport { arcwise } from '@arcwise/control'arcwise.define({tools: ['knowledge.read'],escalation: 'human-review'})
“A model version is not a release until its permissions, evidence, owner, and rollback path are visible together.”
Boundary
Control principle, Arcwise reference model
Explicit scope
Reference control vocabulary
Start with the
control boundary
A concept-stage structure for evaluating fit. No plans are currently sold through this preview.
Inspect
Map one AI system and its release boundary
- System inventory
- Tool and data boundary
- Evaluation record
- Named release owner
- Rollback intent
Govern
Coordinate review across a release path
- Everything in Inspect
- Environment-specific policy
- Reviewer assignment
- Exception register
- Approval checkpoints
- Release history
- Illustrative SDK surface
Operate
Extend the model across internal systems
- Everything in Govern
- Organisation policy model
- Custom release topology
- Connector assessment
- Control ownership map
- Incident context
- Retention requirements
- Implementation workshop
Product concept only. No account, contract, or paid service is created here. Review the control model
Ready to make
release risk visible?
Walk through the reference control model, then decide which policies, evidence, and human gates belong in your own release process.
Product concept · no live account or transaction