Skip to main content
NEES CORE V3 / DEVELOPER PREVIEW

Govern AI
before it acts.

Give your agents capabilities.
Keep their actions accountable.

NEES helps your application check who can do what, on which resource, under which policy—before a supported action proceeds.

A governance boundary for your application.
Your app enforces decisions, executes and verifies.
Developer getting started →

THE ACTION BOUNDARYConceptual

NEES CORE V3

Can this actor take this action?

NEES evaluates the supported request against published authority, policy, scope and governance commitments.

ALLOW ≠ executed. Your application owns the effect.

CAPABILITY IS NOT AUTHORIZATION.Follow an action through NEES ↓

01 / The missing question

It can send the report.
But should it?

ILLUSTRATIVE SCENARIO Not a live result

“Send the confidential
report to this recipient.”

The email tool is connected.
The request still needs authority.

Capability Email tool availableAuthorization Must be evaluated
01

Who is asking?

Is the requester authorized to act?

02

What is in scope?

Are this recipient and this resource permitted?

03

What does policy require?

Can the action proceed, or does it need review?

04

What actually happened?

How will the app verify and report the real outcome?

02 / From intent to evidence

One action.
Four responsibilities.

NEES governs supported requests.
Your application owns execution.
Keep both sides of the boundary explicit.

01

Propose

Describe the action before it happens.

Your app submits a registered action and the context needed to assess it.

Technical detail

Submit

Use the genuine action reference, resource, operation, capability and side-effect attributes. A prompt alone creates no authority.

02

Govern

Ask whether it is authorized.

NEES evaluates published authority, policy, scope and governance state.

Technical detail

Qualify

Use current version-bound assessment commitments. Missing, stale or negative governance must not authorize an effect.

03

Enforce & act

Check the real world, then proceed.

Your application enforces the decision and validates the actual target.

Technical detail

Start → application execution

Proceed only if governance and local context checks permit it. NEES does not automatically execute arbitrary tools.

04

Verify & record

Make the outcome reviewable.

Verify the effect and report what actually happened, including failures.

Technical detail

Report → evidence → receipt

Inspect supported records and reconcile them with application observations. Authorization is not evidence that an action completed.

ALLOW ≠ executed

A decision permits a next step. It does not prove an external effect happened.

03 / The platform, unpacked

Different components.
Explicit boundaries.

Explore the engine, services and developer interfaces that make up the NEES Governance Platform.

SELECT A COMPONENT Conceptual topology

Configuration services / separate from the action flow

Dashboard supports setup and lookup; Control Plane publishes governance
Assessment bindings to your backend · published state for governance

↓ Proposed action + current context

Direct backend clients

Supported mediated paths

↓ Submit supported request

NEES platform / hosted governance

Runtime exposes the lifecycle; Core evaluates governance.

↓ Decision returned · no permission on missing or denied governance

↓ Observe the actual outcome

↓ Truthful report to Runtime · inspect supported records

Read selected component: Hosted Runtime ↓

Application interface

Hosted Runtime

Exposes the governed operation lifecycle to trusted application backends: submit, qualify, start and report, with operation and evidence inspection.

Responsibility boundary

A runtime decision still needs application enforcement and target validation.

Read Hosted Runtime documentation ↗

Select any component to explore its role. Blue boundaries belong to your application; the violet area is hosted governance. Configuration services prepare the context; they are not execution stages.

04 / Put the boundary to the test

Don’t take a diagram
at its word.

Bring a synthetic scenario to Governance Lab. Inspect the facts. Challenge the assumptions. Ask the hosted NEES Core V3 runtime to decide.

Try the Live Governance Lab ↗

Synthetic scenarios. Real hosted governance when available.
No real emails, payments or external actions are executed.

GOVERNANCE LAB / WORKFLOW PREVIEWIllustrative

“An agent wants to share a sensitive report. The requester’s authority is unknown.”
  1. 01
    Describe

    Start with your own synthetic scenario.

  2. 02
    Interpret & inspect

    Review the visible, editable facts. Interpretation makes no verdict.

  3. 03
    Challenge & submit

    Change the facts, then explicitly run hosted NEES.

  4. 04
    Inspect

    Read the returned decision and available evidence.

No predetermined decision. The actual result belongs in the Lab.

05 / Choose your integration

Meet your application
where it works.

Direct from a trusted backend, or through a supported mediated boundary. Select a method to see the setup and responsibilities.

Python SDK / Developer Preview

A direct path for Python agents.

Overview

Connect a trusted Python backend to the hosted Runtime with the public NEESHTTPClient.

Requirements

Python 3.11+, a compatible Runtime, a registered action, a server-side runtime key and current assessment bindings.

Responsibility boundaries

Your service enforces decisions, validates the actual target, executes and reports the real outcome.

Documentation

Open Python SDK documentation ↗

Setup

  1. Resolve the existing assessment in Dashboard.
  2. Install the compatible SDK in your backend.
  3. Use submit → qualify → start → report; inspect evidence and receipt.

Example

Published release candidate · checked 9 Oct 2026

python -m pip install "nees-sdk==3.0.0rc2"

Limitations

Installing the SDK does not register an action or grant authority. Validate API compatibility for your deployment.

06 / Your first governed action

Configure in Dashboard.
Test in your backend.

Free Developer Preview is open: 1,000 new governance submissions per UTC month, one project, one runtime, one development environment, and two active API keys. Production use needs separate deployment-specific validation.

  1. 01

    Set up your workspace

    Verify your email and create one Free organization. Use Developer Onboarding to provision a project, runtime and development environment; store the issued runtime credential on your trusted backend.

  2. 02

    Register the real action

    Define its resource, operation, capability and side effects. Publish required governance through an authorized workflow.

  3. 03

    Resolve existing assessment

    Infrastructure → Assessment configuration manager → Registered action → Published scope → Integration method. Resolve and export the version-bound configuration.

    Read-only lookup. No API key or permission grant. “Publish baseline & resolve” changes governance and may overwrite custom commitments.

  4. 04

    Integrate, exercise, reconcile

    Submit → qualify → start → report. Verify actual outcomes against operation, evidence and receipt.

A guide while you explore

Ask Naina.

Your NEES-governed product and developer-help assistant. Get help understanding the boundary and choosing a path.

Product guidance, with no access to your confidential accounts or permission to change production governance.

Trust / Know the boundaries

Clear about its role.
Honest about its limits.

What NEES does

  • Evaluates supported action requests against published governance.
  • Returns decisions for applications to enforce.
  • Supports review of operation records, evidence and receipts.

What NEES does not do

  • Replace your LLM, agent framework or application.
  • Automatically intercept arbitrary tools or guarantee prevention.
  • Turn an ALLOW or a configuration export into proof of execution.

Build capability. Keep authority.

Your next action
starts with a question.