Skip to content
Menu

AI tools. Human judgment.

Put AI to work.Keep people
in control.

Praxenor is developing tools to check AI permissions and investigate suspicious activity. People decide what is allowed and what the evidence means.

Security products in development. Start today with the free task brief.

AGENT DEFENDER / PERMISSION EXAMPLEIllustration

Writing is not permission to send.

Choose the proposed action
HUMAN REVIEW REQUIREDWITHIN THE EXAMPLE SCOPE

Ask a person first.

The draft stays within the task.

Writing an internal report does not include permission to send it outside the team.

Preparing the internal draft matches the permission in this example. Sending it is a separate action.

Inspect the model

Task: prepare a report. Scope: internal work only. Proposed action: share outside the team.Proposed action: prepare the internal draft.

Inspect the example record
Proposed action
Send outside the teamPrepare an internal draft
Approved scope
Prepare a draft for the team
Decision
Pause for human reviewAllowed in this example
Reason
The recipient is outside the approved workflow.The proposed action matches the approved task.

Illustration only. No agent is connected and nothing is sent. Not evidence of working protection.

Clear scope. Inspectable evidence. Human judgment.About Praxenor

Plan one AI workflow with Praxenor.
Start with a focused review of the task, information, actions and human approvals. Agree the scope and written deliverables before work begins.

See how we can work together →

Veristade / The security product family

Check the action.
Understand the activity.

Veristade is Praxenor’s security product family. Agent Defender asks whether an action is allowed. Swarm Hunter helps explain what activity records show. Choose the job first, then review its evidence and limits.

01 / AI PERMISSIONSIn development

Agent Defender

Check permission before an action.

Designed to compare a requested action with the task’s permissions, ask for required human approval and keep the decision with its reason.

Illustrated permission layers between an internal report and an external send.

The boundary: only tools and actions explicitly connected to its checks are covered. It does not protect every part of a computer.

Explore Agent Defender
02 / SECURITY INVESTIGATIONSIn development

Swarm Hunter

Find the records behind the concern.

Bring records you are allowed to use into one review. Follow related events, check their sources and separate what happened from what remains uncertain.

Illustrated record connections for a fictional security review.

The boundary: a person reviews the findings. It does not automatically block activity or prove who caused it. Accuracy against real threats is not established.

Explore Swarm Hunter
03 / WINDOWS RESEARCHExperimental

Malware Scout

Research repeated patterns in supplied Windows activity records to help a person decide what deserves closer inspection.

Not antivirus or continuous protection. A pattern does not prove malware. No public download.

See the research direction
04 / A PLANNED WORKSPACEDevelopment paused

Stillpoint

A planned AI workspace on your computer for drafts, calculations and reviewing proposed actions.

Development is paused. Not an available service. Privacy and security remain design goals, not proven guarantees.

Read about the concept

Separate products, not one connected system. Any early test needs an agreed task, setup and access.

Compare the details

Something useful, today

A better brief.
A clearer task.

Before you ask AI to work, describe what a good result looks like, which information it may use and when it must ask a person. The free brief helps you put those instructions in writing.

Get the free task brief Build, review and download your own brief. No account required.
A written brief does not enforce tool permissions.
PRAXENOR / TASK BRIEFEXAMPLE

Give the work a clear boundary.

The task
Prepare a report for the team.
The limit
Use only the information approved for this task.
The human decision
Ask before sharing the report with anyone else.

An example of how to describe a task. It does not run an agent or prevent an action.

SEE THE IDEAS IN ACTION

Less guesswork.
Two clear examples.

See how a permission check differs from a security investigation. Press play for narration and animation. Captions and transcripts are available.

AGENT DEFENDER40 seconds · Narrated animation

Check before an AI action.

A report stays internal. A proposed external send needs a separate permission check.

Read the steps and limits

Illustrated scenario with synthetic narration. This is not a recording of live product operation.

  1. 01 / Writing is not sending.An AI tool can write a report. But permission to write is not permission to send.
  2. 02 / Start with a clear task.In this example, the task is simple: prepare a draft for the team. Nothing leaves the company.
  3. 03 / A different action.Now the tool proposes an external send. That request crosses the original boundary.
  4. 04 / Compare the permissions.Agent Defender is designed to compare a connected action with its permissions before it happens.
  5. 05 / A person decides.Here, the send pauses for a person to review. The decision and its reason stay together.
  6. 06 / Know what is covered.This is an illustration, not a live integration. Only actions routed through the checks can be covered.

Use the player’s sound and full screen controls. No agent is connected and no investigation runs from this page.

Open the full transcript
Open the narrated video directly
SWARM HUNTER43 seconds · Narrated animation

Understand suspicious activity.

Follow a fictional trail of failed logins and later requests for sensitive records.

Read the steps and limits

Illustrated scenario with synthetic narration. This is not a recording of live product operation.

  1. 01 / Start with the records.One alert rarely tells the whole story. Start with the activity records you have permission to use.
  2. 02 / Follow what happened.These fictional records show repeated failed logins, followed by requests for sensitive information.
  3. 03 / Connect the sequence.Swarm Hunter is designed to connect related events, so a reviewer can follow the sequence.
  4. 04 / Keep the evidence visible.Each concern should lead back to supporting records. Missing context stays marked as unknown.
  5. 05 / Review, then decide.A person considers normal explanations and decides what to check next. A warning is not a verdict.
  6. 06 / Keep the limits clear.This is an illustration, not a live investigation. The example does not establish accuracy against real threats.

Use the player’s sound and full screen controls. No agent is connected and no investigation runs from this page.

Open the full transcript
Open the narrated video directly

Illustrations, not recordings of working products. No agent is connected and no investigation runs here. Read each example’s steps and limits, or check current product status.

Look beyond the description

See the evidence.
Know the limits.

The October 3 Agent Defender brief describes a small local practice project. A person reviews a specific proposed change, approves or declines it and can inspect the saved result. Read what was checked before deciding whether a test fits your workflow. This does not establish protection across other tools.

Our documentation and internal checks are not an independent security assessment. No certification or proven protection is implied.

Explore the complete permission walkthrough

Compare writing an internal draft with sharing it outside the team. This is an illustration, not a working product.

AGENT DEFENDER / HOW THE IDEA WORKSSame task. Different permissions.

An AI tool can write a report. But is it allowed to send it? Choose an example to see the difference.

EXAMPLE ONLY · NO REAL ACTIONS
Choose the proposed action
Play below, or choose a step

01 / THE REQUEST

What is the AI tool asking to do?

The AI tool was asked to write a report for the team. Now it wants to send that report to someone else.

The AI tool wants to write the report it was asked to prepare. Its actions go through a permission check.

02 / THE PERMISSION CHECK

Check what the AI tool is allowed to do.

What is allowed
Prepare a draft for the team
Proposed action
Send outside the team
Prepare a draft for the team

03 / THE EXAMPLE DECISION

Stop and ask for permission.

This request has permission.

Permission to write a draft does not include sending it outside the team. A person needs to review this request.

Writing the draft is allowed in this example. The check only applies because this tool was set up to use it.

REVIEW REQUIRED · NOTHING IS SENT
EXAMPLE ONLY · REQUEST ALLOWED

04 / THE EXAMPLE REVIEW RECORD

Show the decision and the reason.

Request
External send
Internal draft
What is allowed
Prepare a draft for the team
Example decision
Needs human review
Allowed in this example

Reduced motion is enabled. Choose a numbered step to inspect the story without animation.

These examples show the idea, not the working product. Nothing is connected, sent or approved here.

See what the product covers ↗Inspect the 3D evidence layers ↗

Use it. Inspect it. Discuss it.

A practical place to start.

Free toolkit

Know what your AI tools can access.

List the information each tool can use, whether it can send, change or delete anything, and who approves those actions. Save your inventory as a document.

Create an AI access inventory →

Controlled test results

Inspect a real permission test.

Watch a recording of the results from a local development test. Compare an approved action with refused requests, and see the limits of the test.

See the recorded results →

Work with Praxenor

Turn one workflow into a clear plan.

Discuss the task you want AI to do. Agree a focused review, written boundaries and the checks that would make a useful first test.

Explore the workflow review →

Selected Insights

Questions worth asking before AI acts.

What should happen when an AI action is refused?

A refusal should leave a useful explanation and a clear path to review. Consider what to record before the next attempt.

Read the guide →

Does adding another agent add permission?

Separate the work an agent can delegate from the actions it is allowed to take.

Read the explanation →

How do you investigate suspicious activity?

Compare records and alternative explanations before treating ordinary teamwork as misuse.

Read the review questions →

Browse all Insights → · Choose optional email updates →

Before you start

A few useful
answers.

What can I use today?

The free task brief and AI access inventory are available now. Use the brief to describe one task and the inventory to record what each tool can access and who approves its actions. Both offer editable text and JSON exports. They do not run an agent or enforce permissions. Create a task brief or build an access inventory without an account.

Can I use these products now?

Not yet for general use. The products are at different stages of development. Contact us to discuss whether an early version could be tested on one specific task. Stillpoint remains paused. See how a limited Swarm Hunter test would work.

Do the products work together?

Not as one connected system. Veristade is the family name, and each product has its own job. We can discuss them separately. Using them together would need additional setup and testing.

Does Malware Scout replace my antivirus?

No. Malware Scout is an experimental Windows security companion, not an antivirus replacement. There is no public software download on this website.

See what Malware Scout can and cannot do.

Is Stillpoint available now?

No. Stillpoint development is paused. Its page describes an idea, not an available service.

What should I share in the contact form?

Describe what you need help with and what a useful result would look like. Leave out passwords, access keys, private records, customer information and suspicious files.

We use the details you send to respond to your question. Sending a message doesn’t connect your systems, install software or let us access your files.

What should AI
help you do?

Describe the job, the action you want to control or the activity you need to understand. A useful first step starts with that question.

Contact Praxenor