Skip to content
Khidmitak
Sign in

Enterprise infrastructure for agentic AI

Put AI agents to work in the systems your business runs on.

Khidmitak connects models and agents to your databases, APIs and tools, then governs what they can read, what they can change, and who approves it.

See how it worksTalk to us

A working product. The workspace below runs on synthetic data.

From the demo workspace: a fictional company, synthetic CRM data, real tools, policy and approvals.

Want to put your own questions to it?Request access to the demo

Why now

Can the model answer this?

Can the agent safely do this?

Enterprise AI is moving from answering questions to taking action.

Most organizations already have capable models, and the systems those models need are right there: databases, internal APIs, business applications. Connecting the two takes an afternoon. Letting software act inside them, on behalf of real people and with real permissions, is the hard part.

What it does

What Khidmitak does for your business.

Khidmitak is the layer between AI agents and the systems your business runs on. In practice, that means four things.

Connects everything at once

One answer, from all of your systems.

Most tools answer from the one system they sit on. Khidmitak connects to the systems you already run, from CRM and ERP to accounting, sales and in-house tools, and combines them in a single answer. That's the answer you can't get from any one of them alone.

  • Connect a system and Khidmitak learns its structure. Nobody maps it by hand.
  • Several systems in one workspace, one question.
  • Every figure says which system it came from.
  • CRM
  • ERP
  • Accounting
  • Sales & POS
  • Payments
  • Your in-house system
For example
Which customers with an overdue invoice also have a renewal in negotiation?

A question that only has an answer once your accounting and CRM data meet.

Today Khidmitak connects natively to PostgreSQL databases and to any tool exposed over MCP. For anything else, see 04.

Agentic, not just conversational

It doesn't stop at the answer.

Ask which accounts are slipping, and the agent can prepare what comes next: draft the follow-up tasks, move the deal stage, log the call, in the systems you've connected. People stay in charge of what actually changes.

How a change happens

  1. 1

    The agent prepares

    It drafts the exact change: which records, which fields, in which system.

  2. 2

    You review it

    Nothing has changed yet. The draft shows exactly what will happen, and it expires if no one acts.

  3. 3

    Someone approves

    Only a person whose role allows it. The agent can never approve its own work.

  4. 4

    It happens once

    Khidmitak re-checks the records, then applies the change exactly once. If something changed in the meantime, it stops.

  5. 5

    It's on the record

    Who asked, who approved and what changed is written to an audit trail no one can edit.

That approval step is what makes it safe to say yes to an agent.

Built to not burn money

Predictable cost, by design.

Agent projects get expensive when every question becomes a long chain of model calls, each stuffed with loosely related documents, with more agents running in the background. Khidmitak does the opposite: purpose-built tools ask the exact system for the exact fields a question needs. The model does less work, and every answer stays traceable.

Two ways to answer a question
Two ways to answer a questionOpen-ended agent setupsKhidmitak
Model calls per questionCan chain manyCapped at six
What goes into each callBroadly retrieved documentsOnly the fields the question needs
Between questionsBackground agents watching for changesNothing runs; insights come from rules
Checking an answerHard to trace backEvery figure cites its source

Fewer, smaller model calls mean a bill you can predict, on whichever model you choose.

Backed by engineers who build what's missing

We build the missing connection, and stay responsible for it.

Every enterprise has a system no vendor supports: a proprietary platform, a legacy database, an internal tool. When there's no off-the-shelf connector, our engineers build one for you, maintain it as your systems change, and stay your point of support for it.

  • The new connection gets the same approvals, permissions and audit trail as everything else.
  • We maintain it as your systems evolve.
  • You get a team to call, not a configuration screen to figure out alone.

For technical teams

For the people who check the details.

Your architects and security team will want specifics. The architecture page covers how Khidmitak reaches your systems, how it decides what an agent may do, and what it can never do.

Read the architecture
  • Permissions are checked on the server for every request, per workspace and per role.
  • Database queries written by a model are checked by a real PostgreSQL parser and can only read.
  • Credentials are encrypted, write-only, and never sent to a model or written to a log.
  • Third-party tools stay blocked until an administrator classifies them.

Khidmitak is early-stage software and does not hold compliance certifications today. We're glad to walk any evaluating team through our security model, including its known gaps.

Built in the region

For organizations that decide where their AI runs.

We started in the Middle East, where data residency, Arabic and on-premise systems are everyday requirements, not edge cases.

  • Your infrastructure

    Khidmitak runs as a standard container with PostgreSQL. It isn't tied to a cloud provider, so it can be deployed where your policies require.

  • Your choice of model

    Use any OpenAI-compatible endpoint: a commercial API, a regional provider, or a model you host yourself.

  • Data stays where it lives

    Khidmitak queries your systems in place. It doesn't bulk-copy your records into its own database.

  • Arabic, natively

    The product is fully bilingual, right-to-left included. Ask in Arabic, and work in Arabic.

Company

Who's building Khidmitak

خدمتك

Khidmitak means “at your service”.

That is the standard we hold enterprise AI to: useful inside the organization, and accountable to it.

  • Saif Al Faqih

    Co-founder

    Saif has a background in finance and banking, having worked across fintechs and international banks. He brings the view of the people who run the systems AI is now being asked to act on.

    Saif Al Faqih on LinkedIn

  • Walid Abu Al-Afia

    Co-founder

    Walid is an AI infrastructure engineer. His work centers on compute and infrastructure systems, agent infrastructure and model tooling: what it takes to make models dependable in production, and the safeguards that ensure AI is utilised effectively and executed safely across an enterprise.

    Walid Abu Al-Afia on LinkedIn

Enterprise finance and AI infrastructure engineering: the two sides of the problem Khidmitak exists to solve.

See it working

Don't take our word for it.

The demo is a working Khidmitak workspace for Noor Systems, a fictional company, running on synthetic data. Tell us about your organization, and we'll set you up and walk you through it.

Request accessContact the founders

Every request is reviewed by the founders.

To keep the demo fast, its language model is a scripted stand-in. Every tool call, policy check, approval and audit entry is real.