Vekil · AI agent control plane

Do you know what your AI agents send on your behalf?

What happens if your AI agent sends customer data to a model abroad?

With Vekil: personal data is masked, risky requests come to you for approval, and every step is written to a tamper-evident ledger.

The difference

Without Vekil, with Vekil

Without VekilWith Vekil

No one knows which agent sent what.

Every call is recorded under the agent's name.

Customer data can leave the country as it is.

Personal data is masked; risky requests come to you.

No evidence to show in an audit.

A tamper-evident, signed ledger that auditors verify offline.

app.vekil.tech / Overview
Vekil Overview screen: calls, denied, awaiting approval and cost cards; the agent fleet table with status, risk, calls, cost and monthly budget
Overview. Calls, denials, approvals and cost; every agent's status, risk and budget on one screen. All figures are computed from the ledger.Screenshot made with sample data.
Control and compliance

What Vekil does

Data turning into anonymous tokens behind a shield

Masks.

National ID numbers, cards, phone numbers and emails reach the model or external system masked. Personal data never leaves for an unregistered foreign destination. Requests containing an API key, a password or a .env file are not sent at all.

KVKK Art. 9 · GDPR
evidence ledger · event #41207sample record
request
TC 1••••••••0 → <TR_TCKN_1>
target
model · abroad · sent masked
policy
ALLOW · if unmasked ASK
chain
9f3a…c1e0 ← 7b20…e4d9
event #41208
request
.env line · DB_PASSWORD=••••••••
target
model · not sent
policy
DENY · forbid_secret_leak
Illustrative example; not a real record.
An approval symbol at the centre of interlocking light rings

Asks.

Requests the policy finds risky wait for human approval; the decision can also be made from Teams or Telegram. Any agent or model can be stopped with one click.

Human approval · kill switch
Linked, glowing blocks stacked layer on layer

Proves.

Every call is written to a hash-chained, append-only ledger and signed at the end of the day. Auditors verify the record offline.

Tamper-evident ledger
app.vekil.tech / Live events
Live events table: model and tool calls, a policy denial, an A2A call and masked personal data findings; next to it, the ledger integrity card showing the signed daily seal
Live events and ledger integrity. Masked data, policy denials and the day's signed, timestamped seal.Screenshot made with sample data.

Data masking aligned with KVKK Art. 9, an instant kill switch and a cryptographic audit trail you can verify offline.

Enterprise chat

Vekil Chat

Your employees already use AI, most of them with accounts they found themselves. Vekil Chat gives them the company's own chat screen. Every message passes through Vekil's gate.

  • Sign in with Microsoft. The account is created on the first sign-in from your company domain.
  • Personal data is masked. National IDs, IBANs, phone and card numbers and secret keys are masked before they reach the model; employees see what was masked.
  • Policy applies to chat too. A message that hits a rule is not sent, or waits for a manager's approval.
  • Attachments are inspected. PDF, Word and Excel files are converted to text and go through the same inspection; the file itself is not stored.
  • Who, when, which model. Every call is written to the append-only ledger with a verified identity; you choose the models and the monthly budget.
  • Encrypted history. Deleted when the retention period ends; an erasure request deletes the person's chats too.
  • Your brand. The screen opens with your logo and colours.
Personal account

Write a refund letter for customer Ayşe Y., ID 10000000146, IBAN TR33 0006 1005…

Goes to the model as is. No rule, no approval, no record.

Vekil Chat

Write a refund letter for customer Ayşe Y., ID <TR_TCKN_1>, IBAN <TR_IBAN_1>

2 personal data items sent masked (national ID, IBAN)

Sample values are synthetic. Free-text details such as names may not be recognised.

In the Professional and Enterprise editions, with no limit on users.

Live example

Try it yourself

Vekil inspects the traffic of the agents connected to it. Edit the text below; the masking updates as you type. Add an API key or a DB_PASSWORD=… line and see what happens.

What reaches the model
This request is NEVER sent to the model: it contains a secret (API key, password or private key). Vekil rejects it and writes the denial to the ledger.

What would reach the model without the secret:
Summarise this customer refund request:
ID <TR_TCKN_1>, phone <TR_PHONE_1>
Email: <EMAIL_1>
Refund IBAN: <TR_IBAN_1>
Card: <CREDIT_CARD_1>
DB_PASSWORD=<SECRET_1>
  • Turkish ID · 1
  • Phone · 1
  • Email · 1
  • IBAN · 1
  • Card · 1
  • Secret (key, password) · 1

This box runs in your browser; nothing you type is sent anywhere. It is a small sample of Vekil's full recognizer.

Videos

Watch it at work

One-minute videos recorded in the real product with demo data. All are subtitled, so they work with the sound off.

How Vekil works0:58

A one-minute animation: agents, the gate, masking, human approval, the evidence ledger and Vekil Chat.

Overview and fleet0:59

Every number on the dashboard is computed from the evidence ledger.

Personal data masking0:53

National ID, phone and email reach the model masked.

Human approval and the kill switch0:47

The agent waits for approval; a misbehaving agent stops within a second.

All videos (19)

Architecture

One gate between your agents and the world

Your applications connect to Vekil with their existing SDKs. Model calls, API requests and MCP tools all pass the same checks.

Vekil architecture: assistants, automations, workflows and partner agents pass through Vekil as a single gate, where masking, policy, human approval, the secret vault, the kill switch and the evidence ledger apply, before reaching models, business systems and MCP servers.
Yours

Agents

  • Assistants and bots
  • Automations
  • Scheduled workflows
  • Partner agents (A2A)
Vekil

Control plane

  • identity and budget
  • personal data masking
  • policy and human approval
  • secret vault
  • kill switch
  • evidence ledger
Outside world

Models and systems

  • OpenAI, Anthropic, Azure
  • ERP, CRM, marketplaces
  • MCP servers
Licensing and scale

Three editions, the same protection

Editions differ in the number of agents and in compliance, reporting and integration features.

Starter

Community

3agent
  • The full protection layer
  • Evidence ledger and verifier
  • Overview and fleet dashboard
  • Logs and support bundle
Recommended

Professional

10agents + extra agents
  • Evidence package (for auditors)
  • Data subject requests (DSAR), RoPA, DPIA
  • Single sign-on (SSO)
  • Teams / Telegram approval notifications
  • Visual workflows
  • Vekil Chat, unlimited users
Scale

Enterprise

50agents + extra agents
  • Everything in Professional
  • SIEM export (syslog / JSONL)
  • RFC 3161 timestamping

Protection is on in every edition. Masking, policy, the transfer gate, human approval, the kill switch and the evidence ledger work regardless of license. Traffic keeps flowing even if a license expires.

Infrastructure

On your server, with your data

  • On your own infrastructureInstalls on a single server with Docker; records and secrets stay in your database.
  • Offline license checkThe license is verified without going online; usage data never reaches us without your consent.
  • Audit-readyThe evidence ledger is signed daily; auditors verify records with a separate tool, offline.
  • Turkish, English, GermanThe admin interface and server messages come in three languages.
A locked server cabinet wrapped in light ribbons
Contact and demo

Let's talk demo and pricing

We'll show you how it works with your own agents and your own data.

Direct contacthello@lucoragroup.com