A boundary, not a brochure.
Constraint Native is a narrow, local, in-process boundary. This page describes what it does, what it does not do, and how a reviewer should evaluate it. The claims are bounded; the non-claims list is explicit so the scope is unambiguous.
Hierarchy: Golem governs what can be said; Constraint Native governs what an agent can do.
What is in scope.
What is out.
The current scope is intentionally narrow. Hardening, second-platform alpha, and external review are funded work — not pre-funding promises.
In scope
- Local Agent Firewall + MCP Gateway, in-process with the coding agent
- File / shell / MCP / network governance with deny-by-default posture
- Capability brokering, taint tracking, approval routing, canary-secret blocks
- Ed25519-chained signed event log, replay-audit walkthrough
- macOS alpha · single-developer workspace
- Bounded blast-radius reporting per session
Out of scope
- Compliance certification claims (SOC 2, FedRAMP, HIPAA, ISO)
- Endpoint isolation, sandbox containment, hypervisor-grade boundaries
- Replacement for IAM, EDR, DLP, secure code review, or human judgment
- Hosted multi-tenant gateway (we are local-first and in-process)
- Perfect prompt-injection prevention guarantees
- Cross-platform GA — Linux/Windows alphas come after the bridge proof is reproducible
What we model.
How we contain it.
How a reviewer can check.
The boundary is small enough that a reviewer can hold the entire model in their head. Each row maps a concrete threat to a concrete control to a concrete piece of evidence in the proof artifact.
Local-first by default.
Telemetry off by default.
The proof artifact stays on the developer's machine until they choose to export it. There is no cloud control plane required for the boundary to work.
What to verify in 30 minutes.
If you are reviewing the boundary, these are the seven checks that matter most. Anything that surprises you here is something we want to know about.
Trust starts with the unfinished edges.
These are planned diligence items, stated before a reviewer has to ask for them.
Before external review
- Alpha currently scoped to macOS and single-developer workspace.
- Full containment requires Constraint Native to own the launch boundary.
- External audit is not complete yet.
- Cross-platform behavior is not yet validated.
- Fresh live proof export should be generated before diligence.
Stated explicitly.
Constraint Native is bounded by design. The lines below are the ones we deliberately do not cross — at least, not yet, and not without external review.
Constraint Native is not
- Compliance certification (SOC 2, FedRAMP, HIPAA, ISO 27001, etc.)
- Full endpoint or hypervisor-level isolation
- A perfect prompt-injection containment guarantee
- A replacement for IAM, EDR, DLP, secure-code-review, or human judgment
- A peer-reviewed or third-party-audited security product (yet)
- A cross-platform GA — alpha-scope until external review and second-platform port
- A clinical, medical, or consciousness-related product
Want to pressure-test the boundary?
Engaging an external reviewer is a funded line item in this round. If you have threat-modeling or security-review experience and want to look closely, we want to hear from you.