NAHUM

01 — An AI agent for one company at a time

An AI agent that works
for one company: yours.

Local. Model-agnostic. Yours to keep.

Talk to us →

02 — Why it's different

Local. Nahum runs on your machine, or on your own server. Your data doesn't leave the building.

Model-agnostic. OpenAI, Anthropic, DeepSeek, or a model running on your own hardware — switch any time. No dependency on one provider.

More versatile, more flexible. It isn't a chatbot. It uses real tools: a terminal, files, a browser, your desktop, email, messaging — Telegram, WhatsApp — scheduled tasks, and parallel subagents when a job calls for more than one hand.

A company brain. Persistent memory and skills. It knows the people and the business — clients, suppliers, prices, processes, preferences, history. It doesn't start from zero every morning.

Continuous learning. It builds skills out of experience, sharpens them with use, and keeps a deeper model of who you are the longer it runs.

03 — What a real agent does

Alcendo — Berlin, German accounting, DATEV

A bookkeeper that never sleeps on receipts.

Reads scanned invoices, names them to the firm's own filing scheme, reconciles them against bank statements, and exports the matches to DATEV — reporting the result, one invoice or twenty, as a single message. It also builds the firm's investor decks: a financial model in, a finished PDF and an editable PPTX out.

Your own paperwork, handled the way your firm already does it — not retrained from a template.

Bitdriver — El Salvador, taxi dispatch

Dispatch that runs itself, one WhatsApp message at a time.

Orders arrive over WhatsApp. The agent greets the customer, gathers the details — pickup, drop-off, passengers, luggage, flight — posts the job to the driver group first-come-first-served, confirms the fare and a photo of the car once a driver claims it, and books the ride.

Whatever runs on your phone right now, by hand, can run like this instead.

Canopi — Portugal, land & location intelligence

A report that writes itself, checked before it ships.

Turns an address into a quality-of-life report — air, noise, water, green space, nearby services, travel times, natural risks — pulled from public data, rendered as a branded PDF with a public link, and held to an automated quality gate before a client ever sees it.

If your business sells something that used to take a person a day to assemble, this is what it looks like done in minutes, the same way, every time.

Case 04 — in preparation

A real trace, abridged

> invoice intake: 3 new PDFs found in shared drive
> naming: renamed per filing scheme (2026-10-09_Supplier_Amount_v1.pdf, x3)
> reconciliation: matched 2 of 3 against this week's bank statement
> export: moved 2 matched invoices to DATEV/2026/10/
> status report -> Telegram #General: "2 invoices matched and exported, 1 waiting on a statement"

04 — Who owns it

The client is the owner.

You own the agent, its data, its prompts, and its repository. No seat licenses, no rented intelligence, no black box. If you leave tomorrow, you take all of it with you.

05 — How you get one

  1. Install it on your machine, or on a server we provide.
  2. Connect it to Telegram, WhatsApp, or email — wherever your work already happens.
  3. Teach it your company: the people, the processes, the history it needs to know.

FAQ

Does our data ever leave our servers? No. Nahum runs where you run it — your machine or your server, your choice of model provider.

What if we switch AI providers later? You switch. Nahum isn't built around one model, so there's nothing to migrate away from.

What if we want to leave? You already own the repository. There's nothing to hand back.

Contact

Tell us what you'd want it to do first.