PSPriya Shah, finance analyst

My invoice checker runs on my laptop. Half of finance asks me to run it for them.

Your AI enthusiasts build the agents. The whole company uses them.

Start a pilotInside your perimeter, under your rules.

donkit.acme.internal/chat/invoice-checkLive

% python check.py inbox/

reading 14 invoices…

INV-2291 unit price 14.20 != 13.80 §4.2

INV-2291 terms 30d != 45d §6.1

INV-2297 ok

INV-2304 ok

saved results_0912.csv

Works. For one person.

Runs only on Priya’s laptop

Finance emails her the PDFs to check

Billed to her personal API key

Invoice CheckMy settings

INV-2291 · Northwind Supplieschecked against contract C-114

2 mismatches

FieldInvoiceContract

Unit price€14.20€13.80 · §4.2

Payment terms30 days45 days · §6.1

Quantity1,2001,200

VAT20%20%

Overcharged on this invoice€480

1,200 × €0.40

Send to supplierApprove anyway

JRAKMWSO

38 people in Finance use itGuard · Enforce

(01)Where agents live today

Your best agents live on laptops.

Every company already has people who build with AI: an invoice checker, a handbook bot, an NDA reviewer. They work. But they run on one person’s machine, with that person’s keys, and nobody else can use them.

On the author’s laptop

Who can use itThe author

When it worksWhile the laptop is open

Whose accounts it usesThe author’s, for everyone

Who paysThe author’s own card

Who sees what it doesNobody

On Donkit, inside your perimeter

Who can use itEveryone you publish it to

When it worksAlways, on your servers

Whose accounts it usesEach colleague connects their own

Who paysOne company balance, a limit per person

Who sees what it doesIT, down to every decision

(02)Who builds

Builders don’t need to be engineers.

Builders are the people in every team who already got good at AI: the analyst with a folder of prompts, the HR partner with a handbook bot, the lawyer who checks NDAs with a chatbot. They know the work. Donkit does the engineering.

The builder brings

How the job is done, the exceptions, the documents and a few real examples.

Donkit brings

The prompt, tools and connections. Then hosting, access and rules for everyone who uses it.

Editing @ Invoice CheckYou are in builder mode — not the live use chat.

Exit edit

Builder chat

Check each supplier invoice against its contract. Flag unit price, payment terms and VAT. The whole accounts payable team will use it.

Writing the system prompt· 3 steps

Reading contracts on the Finance drive

Enabling read_pdf, drive_search

Invoice Check is ready. For each invoice it finds the supplier’s contract on the Finance drive and compares unit price, payment terms and VAT. Every mismatch comes with the clause it breaks.Try it on a real invoice before you publish. Should a mismatch go back to the supplier right away, or wait for someone’s approval?

Wait for approval from the AP lead

Auto

(04)Safety

IT writes the rules in plain words.

Admins describe what agents may and may not do, one thought per line. Donkit turns that into rules, shows each one before saving and applies them to every agent in the organization within half a minute. Your rules only add limits: the stricter one always wins.

SecurityHow the guard treats what your agents send to models and what they do.

Guard mode

Secrets are masked, personal data is masked before outside models, side effects ask you first.Applies to every agent within about half a minute

Describe what your agents may and may not doWrite it in your own words, one thought per line. We turn it into rules and show you each one before anything is saved.

Nobody may delete files.Sending email always asks the person first.No customer data to models outside the platform.

Draft rules3 rules drafted

Organization rules3 rule(s)

blocksBlocks the tools delete_file for everyone.

asksAsks the person to confirm the tools send_email.

masksMasks what is found as personal data in messages to models outside the platform.

Save as version 4Version 3 stays one click away.

DecisionsThe newest decisions of the guard for this account.

14:02

NDA Reviewl.ortiz · personal data → outside model

Masked

13:58

Invoice Checka.kim · send_email

Asked for approval

13:41

Onboarding Buddyj.reyes · delete_file

Blocked

13:40

Deal Desks.oduya · crm.read

Allowed

Version historyv3 · activeRoll back

(05)Where they live now

Now they live inside your perimeter.

The same Donkit, installed in your cloud or on your own servers. Agents call your models, use your systems and follow your rules. Nothing has to leave.

Your cloud or your serversRuns on your Kubernetes, installed with Helm. Or in our cloud, if that suits you better.

Installs offlineEverything ships as one archive. Nothing is downloaded during installation.

Your modelsAzure OpenAI, Vertex, Anthropic, Ollama, vLLM or any OpenAI-compatible endpoint. Switch platform models off and only yours remain.

No outside servicesBilling, system email and the skills marketplace stay off. People sign in with internal addresses.

Bring your agents in from the laptops.

Start a pilotStart with one team and the builders it already has.