Capability statement

CoDeS LLC · November 2026 · Governance Before Action

Governed AI · Cybersecurity R&D · Local-first software systems.

The government problem

As agencies introduce AI assistants that can change files, permissions or workflows, they need to know who approved an action, what the AI actually did, and whether the result was verified. That is especially important for high-accountability and intermittently connected systems.

What CoDeS does

CoDeS LLC is developing local-first software that places permission checks between an AI recommendation and an action. It can hold unauthorized requests, record governed work, and show reviewers the evidence. This is an internally tested R&D prototype available for technical evaluation and scoped demonstration. It is not a certified production product.

Core technical capabilities

• Pre-action controls: check authority before AI tools change files, permissions or systems.
• Reviewable evidence: link a request, identities, policy decision, execution and independent observation in a job graph.
• Local-first operation: demonstrate governed workflows on operator-controlled Android and Windows hardware.
• Honest recovery: distinguish blocked work, stale evidence and missing information from verified success.

A simple use case

An AI assistant proposes changing system permissions. CoDeS checks the exact request and authority before proceeding. If approval is missing, the request is held. When an action is authorized and performed, a reviewer can inspect the recorded decision and the observed result.

Evidence from internal product R&D

• A bounded October 17 job-graph test returned 13 nodes and 18 typed relationships: 7 hash-matched, 11 owner-recorded and 0 mismatches in that sample.
• An October 19 code-location benchmark found 60/60 correct files and lines, and 27/27 invented or out-of-scope symbols were correctly reported as not found. No model was used for those questions.
• Governed development jobs have produced execution, Proof and UNIFY records. These results are internal R&D evidence, not independent product certification.

Read the technical evidence summary

Why this approach reduces risk

AI ability is not permission to act. Reviewers can examine the evidence rather than rely on an AI-generated “done” message. Missing or stale data remains visible and cannot silently become proof of success.

How an organization can engage

We are seeking R&D collaboration, scoped technical demonstrations, feasibility assessments and controlled pilot design. Production deployment, licensing and independent validation would require further qualification and agreement.

Company data

Assurance note

R&D is informed by NIST AI RMF concepts. No NIST endorsement, certification, federal contract past performance, or independently certified product is claimed. Additional redacted evidence is available for qualified technical review.

Download the PDF