SENTINEL AI

LOCAL-FIRST AGENT SECURITY

Your agent can propose.
Only policy can permit.

Put an independent security gateway between an AI agent and its tools. Keep task authority on your machine, even when the instructions it reads cannot be trusted.

Get Sentinel AI ↗Read the documentation

Rust enforcement · Python integration · Optional cloud visibility

THE AUTHORIZATION BOUNDARY 01 / LOCAL
01
AI agentUntrusted proposal
read_file
↓
02
Sentinel gatewayIdentity · Scope · Expiry · Policy
TRUSTED
ALLOWDENYAPPROVAL
↓
03
Restricted toolExecute only after authorization

Illustrative request flow. Every supported action receives its own decision and audit record.

AUTHORIZATION IS NOT A MODEL OUTPUT

Useful autonomy needs
a separate source of authority.

A malicious document can influence a model. It cannot sign a task contract, grant an approval or expand the gateway’s permissions. Sentinel enforces those decisions in trusted code outside the model.

Small permissions.
Clear boundaries.

Explore the controls ↗
01 / AUTHORIZE

Bound every task

Signed contracts describe which operations and exact resources a task may use, and when that authority expires.

02 / ENFORCE

Check every request

Default-deny policy, one-time approvals and restricted execution keep decisions separate from model-generated instructions.

03 / OBSERVE

Understand the outcome

Inspect local audit records. Optionally synchronize an allowlist of event metadata to a private account.

YOUR MACHINE IS THE CONTROL PLANE

Cloud visibility.
Local authority.

The online dashboard observes device-reported events. It cannot run commands, read files, approve actions or edit trusted policies on your computer.

Synchronization is off by default. Prompts, documents, tool responses, resource names and signing keys stay local.

Read the privacy architecture ↗

Start with a controlled task.

Getting started ↗
  1. Install a verified package

    Choose a supported platform and verify its checksum before installation.

  2. Connect supported tools

    Use the Python SDK to route every protected operation through the Rust gateway.

  3. Inspect real decisions

    Run synthetic allowed and denied requests; inspect their local audit records.

Sentinel does not automatically protect arbitrary applications. Tools that bypass its gateway are outside its security boundary. Understand the limitations.