Trust kernel for agents2026Active

Sentinel OS

A local-first trust kernel for agent systems with policy checks, append-only history, and verification.

Back to projectsRepository

At a glance

Problem

Serious agents need enforceable boundaries and a trustworthy record of what they actually did.

Approach

A local-first trust kernel with policy checks, append-only history, and verifiable outcomes.

Why it matters

Moves agent safety away from vague logging and toward systems-level integrity guarantees.

System sketch

01Constrain

Policies are evaluated before actions run, with a bias toward explicit permission rather than best effort.

02Record

Execution history is captured as an append-only sequence of transitions instead of scattered logs.

03Verify

Outcomes are packaged so that integrity can be checked offline without trusting the running agent.

04Extend

The structure is designed to grow toward auditable memory and tighter control planes for agents.

Design notes

Trust must be reconstructable

If a system cannot show what happened after the fact, it is not trustworthy enough.

Policies should be visible

Constraint systems matter more when humans can see the rules that governed an action.

Integrity is infrastructure

Verification belongs in the system design, not as an afterthought.

Question

The harder question for serious agent use is not raw capability. It is whether actions can be constrained, reconstructed, and verified without trusting the model on its own terms.

Sentinel OS treats that as a systems problem rather than a logging or prompting problem.

Approach

  • Use append-only execution history instead of scattered logs.
  • Bias policy evaluation toward explicit permission rather than best effort.
  • Package outcomes so they can be verified offline after execution.
  • Treat memory and state transitions as objects that should eventually be auditable too.

Open questions

  • What is the smallest useful policy surface for practical agent systems?
  • How should memory mutations be represented if they need the same audit trail as actions?
  • Which guarantees are worth paying for in exchange for complexity and friction?

Project spec

RoleConcept, system design, and implementation.
StatusActive
TypeTrust kernel for agents
UpdatedFeb 2026
Primary objectSigned policy decisions and append-only action records.
System boundaryPre-action constraint checks and post-action verification.
Current artifactLocal-first trust-kernel prototype.
Pressure pointExtending the same audit model to memory and state mutation.

Topics

trustworthy agentspolicy enforcementcryptographic verificationexecution integrity

Where it is now

  • The current public milestone covers signed policy decisions, append-only records, and basic verification.
  • The next useful step is replay, richer inspection tooling, and a clearer model for memory and state transitions.