TillarTechnologies

Bind every AI action to the authority that allowed it.

Deterministic execution control that binds proposal, authority, exact permitted effect, and independently observed outcome into one reconstructable record.

Open the Evidence Lab Discuss a design partnership
Recorded fixture run Refusal captured
Public set
7 files
Negative test
Visible failure
Scope
Fixture-only

Development-operational, fixture-scoped

The current Linux fixture has exercised typed intake, signed authority checks, revocation verification, a durable one-use decision, one bounded scratch-file effect, independent observation, and reconciliation. Public downloadable evidence remains the narrower refusal-and-custody demonstration. Neither is production or off-host validation.

No layer silently promotes itself into authority.

A proposal is not permission. Supporting evidence is not authorization. A receipt is not proof of truth. CageOS is designed to preserve those distinctions in the execution path, so an agent's confidence, explanation, or tool call can never become its own approval.

  1. 1ProposalWhat an AI or operator recommends
  2. 2EvidenceWhat supports or challenges it
  3. 3AuthorizationWho may approve this action
  4. 4ExecutionWhat actually occurred
  5. 5ObservationWhat an independent witness saw

One action record. Different questions for every reviewer.

CageOS is designed for teams that must let coding, infrastructure, or operations agents propose state-changing work while preserving defensible authority, evidence, and reconstruction.

AI and platform teams Did this agent action stay inside the approved task? Examine the proposal, exact scope, controlling authority, effect, and observed result.
Security and model risk Could approval for one action be replayed or widened? Inspect signed bindings, expiry, revocation, single-use handling, and refusal records.
Audit and compliance Can we reconstruct who allowed what and what actually changed? Follow one action-specific chain without treating an agent trace as authority or outcome proof.
Federal and regulated buyers Can acceptance criteria map to inspectable artifacts? Test bounded controls and evidence without depending on inaccessible vendor-only assertions.

One foundation, not three parallel products.

CageOS is the foundation. CageOS Lite is planned to demonstrate its complete authority path in a lower-stakes setting. PredX is planned to apply the validated foundation to preclinical evaluation and portfolio triage.

CageOS

Core platform. Current priority.

An evidence-producing authority boundary for AI-enabled operations.

  • Explicit, typed authority
  • Fail-closed execution controls
  • Inspectable evidence and custody
Explore CageOS

CageOS Lite

v0.1.0 published and signed

A signed, independently verifiable demonstration: run a real fixture-only lifecycle and check its claims against reality yourself.

Get CageOS Lite v0.1.0

PredX

Planned first application

Governed preclinical evaluation and reconstructable portfolio triage, with negative results preserved and claims bounded by their support.

Explore PredX

Bind the seams between agent, identity, change, and audit systems.

CageOS is intended to integrate with agent frameworks, Git and CI/CD, IAM or PAM, ticketing, and SIEM, not replace them. Its proposed advantage is the action-specific chain those separate systems do not necessarily create on their own.

This describes architectural alignment, not ATO, cATO, certification, or agency approval.

OMB M-25-22 DoD cATO evaluation criteria NIST AI Resource Center
Agent frameworks
Receive proposals without granting the model adjudicative authority.
Git and CI/CD
Bind an approved change to the exact effect and retained outcome record.
IAM and ticketing
Carry named authority and scope into the execution path instead of leaving them adjacent to it.
SIEM and audit
Export reconstructable refusal, execution, and observation artifacts for review.

Inspect a real refusal, not an animation.

A fixture-only CageOS Edge refused ungovernable input, retained the submitted bytes, recorded the event, and detected an altered copy. Download the frozen bundle and run the checks without installing CageOS.

Open the Evidence Lab Download the bundle

Evidence before assertion.

Tillar Technologies LLC is a disabled veteran-owned Florida company developing deterministic authority and evidence infrastructure for regulated, federal, scientific, and other high-consequence AI work.

Bring one bounded coding, infrastructure, or operations use case, the authority governing it, and the condition that must cause refusal. We can test whether the Change Passport approach is worth its operational cost.

Discuss a design partnership
Company
Tillar Technologies LLC, Florida
Ownership
Disabled veteran-owned
Commercial wedge
AI Change Passport
Current platform
CageOS fixture implementation
Public evidence
Recomputable custody demo