Custom AI agents that do real work inside your systems
An AI agent reads, checks and acts in your email, CRM, WhatsApp and spreadsheets: it looks up a new lead, reads a supplier invoice in Hebrew, notices a missing document. You decide what it does on its own and what waits for your approval.
- Built in Netanya, supported after launch
- Every action logged, the important ones approved by you
- Works with Hebrew, English and Russian text
What is an AI agent for business, and what do we build?
An AI agent is software built on a language model that doesn’t just answer questions: it carries out a task inside your tools. It reads an email or a PDF, looks things up on the web or in your CRM, picks the next step by your rules and writes the result where your team works. We build custom AI agents for businesses in Israel, with limited permissions, human approval for anything sensitive and a log of every action.
Does this sound like your week?
Every new lead starts with a round of Googling
Someone opens the company’s website, checks the registry and LinkedIn, then copies the details into the CRM by hand. By the time the lead is qualified, a faster competitor has already called back.
Supplier invoices pile up in the inbox
PDFs in Hebrew from different suppliers, each with its own layout. At the end of the month someone types them into Excel, and a quiet price increase only gets noticed when the accountant asks about it.
ChatGPT helps, but only inside its own window
Your team already asks AI for drafts, then pastes the answer into WhatsApp, the CRM and the spreadsheet. The AI never sees your data and never finishes the job, so the copy-paste has simply moved somewhere new.
Nobody wants AI talking to customers unchecked
One wrong message to a client or one deleted record costs more than all the hours it saved. So the AI pilot stays a demo, and the work stays manual.
How we build an agent, step by step
Pick one task
We start with a single repetitive task that eats your team’s time and agree on what a good result looks like. If a simple automation or a ready-made tool will do the job, we’ll say so.
Watch how it’s done today
We sit with the person who does the work now: where the inputs come from, which decisions they make, which exceptions keep showing up. That becomes the agent’s rulebook.
Build it with guardrails
The agent gets only the tools and permissions the task needs: read here, write there, never delete. Anything sensitive needs a person’s approval.
Run it in suggest mode
At first the agent prepares and a person approves or corrects every result. We go through the log together and tune the rules on your real cases, not on made-up examples.
Let it act where it has earned trust
Steps that come out right consistently switch to acting on their own; the rest stay under approval. Then we add the next task.
What you get
- A task map: what the agent does, what stays with people, and where it must stop and ask
- A working agent with its own tools, rules and least-privilege access to your systems
- Connections to your email, CRM, WhatsApp, Google Sheets or Drive, through APIs or an MCP server
- An approval flow: drafts reach you on WhatsApp or by email, ready to approve, edit or reject
- A readable action log: what the agent did, based on which source, and why
- Usage limits and alerts, so model costs don’t creep up unnoticed
- Monitoring, tuning and support after launch
Tasks an agent can take off your plate
Qualifying inbound leads for a B2B supplier
A request comes in through the website form. The agent checks the company’s site and public registry details, fills in the CRM fields, scores the fit against your criteria and drafts a reply in the lead’s language. A salesperson approves and sends.
Supplier invoices for a restaurant or a contractor
The agent collects invoices from the mailbox, reads the supplier, date, amount and VAT, matches them to orders and adds a row to the bookkeeper’s sheet. Price changes and mismatches are flagged, never quietly fixed. The summary can reach you in English even when every invoice is in Hebrew.
Chasing documents for an accounting or insurance office
Clients send forms and payslips on WhatsApp and by email. The agent checks what has arrived against each client’s checklist, spots a blurry or wrong page and drafts a polite reminder in Hebrew, English or Russian for your team to send.
Drafting quotes from WhatsApp requests at a building-supplies store
A contractor sends a shopping list or a photo of a handwritten order on WhatsApp. The agent matches each item to your price list, checks that customer’s agreed prices in past orders and prepares a draft quote. Your team reviews it, adjusts and sends it, so the contractor isn’t left waiting until someone has a free evening.
Research and data enrichment agents: complete lead records, with sources
A research agent takes a name, a company or a website and fills in what’s missing: industry, signs of company size, contact channels the company itself has published, recent news and how well it fits your ideal customer. Next to every field sits a link to where it was found, so a person can check it before it influences a decision. The same approach works for a weekly watch on competitors’ public prices and product range. This is data enrichment and web research done carefully, not bulk scraping.
- Public sources and official APIs only: the company’s own site, public registries, search results
- A source link next to every field, and “not found” instead of a guess
- Human review before enriched data drives a decision or a message
- We respect site terms and Israel’s Privacy Protection Law: no personal data without a legal basis, no scraping behind logins or around blocks
- Outreach only to people who agreed to receive it: Israel’s anti-spam rule (section 30A of the Communications Law) applies to marketing by email, SMS and WhatsApp
How an agent stays under your control
An agent is only as safe as the limits around it, so we design the limits first and the clever part second.
- A separate account for the agent, with access only to the folders, fields and channels its task needs
- Irreversible actions, such as messaging a customer, paying or deleting, always wait for a person
- Keys stay on the server, never inside prompts, and you can revoke them at any moment
- A handoff to a person whenever a case is unclear or data is missing
- A spending cap per task, alerts when a step fails, and a way to pause the agent at any time
We run our own product on agents
NetzFlights, our site with flight deals from Israel, works this way: agents search for fares and draft the post for each deal, and a person checks every deal by hand before it goes out. Two or three deals a day, with dates, airline and baggage. It’s the same split we build for clients: the agent does the legwork, a human keeps the final word.
NetzFlights
Flight dealsFlight deals from Israel: two or three a day, each checked by hand. Only fares that genuinely beat the usual — with dates, airline and baggage.
- An easy first step
Name one task your team dreads every week, and we’ll tell you honestly whether an agent can take it over.
Tools an agent can work with
- Gmail
- Outlook
- WhatsApp Business API
- Google Sheets
- Google Drive
- Google Calendar
- CRM via API
- n8n
- MCP
- Claude API
- OpenAI API
If a system has an API, we can usually connect it. If it doesn't, we'll tell you upfront.
How a project with us runs
The same four steps, whether it's a bot, an integration or a whole system.
Short call
You tell us what's slowing you down and how you work today. We ask questions — and if a ready-made tool fits, we'll say so.
Your partDescribe the task in your own words
Written proposal
Scope, stages and a cost estimate in writing, including who owns the code, the accounts and the data.
Your partReview, ask questions, decide
Build in stages
You see working parts early and try them on real cases. Your feedback goes straight into the next stage.
Your partTest and give feedback
Launch and support
We launch, watch how it behaves in real use and stay on for fixes and improvements.
Your partUse it and tell us what to improve
What affects cost and timeline
- How many steps, decisions and exceptions the task involves
- How many systems need connecting, and whether they have an API
- Where the agent only reads and where it also writes: every write needs approvals and testing
- Volume: how many documents or requests a day or a month, which sets the running cost
- The state of your inputs: scans, WhatsApp photos, handwriting, several languages
- How much reporting and logging you need
- The level of support you want after launch
An agent has two costs: the one-time build, and the running cost per use (model and API fees paid to the providers). We estimate both in writing after a short call. Call or WhatsApp us on 050-854-2579.
Ownership, privacy and accessibility
Your code and data
Ownership of the code and data is agreed in writing before we start: who owns the source code, the accounts and the information is in the agreement — not something you find out at the end.
Privacy by design
We build with Israel's Privacy Protection Law and Amendment 13 in mind: we collect only what's needed, keep secrets on the server and agree with you where the data is stored.
Accessibility built in
Interfaces built with IS 5568 and WCAG AA in mind: keyboard navigation, contrast, screen-reader labels and right-to-left layouts.
Questions about AI agents
How is an AI agent different from a chatbot or a regular automation?
A chatbot answers questions in a conversation. A regular automation follows a fixed path: when X happens, do Y. An agent sits between them: it gets a goal, reads messy inputs such as emails, PDFs and web pages, chooses the next step within your rules and uses your tools to finish the task. When a fixed automation is enough, we’ll recommend that instead, because it’s simpler and cheaper to run.
What happens when the agent gets it wrong or doesn’t know?
Language models do make mistakes, so we plan for it. The agent starts in suggest mode, where a person approves every result. It’s instructed to write “not found” rather than guess, and to pass the case to a person when data is missing or the rules don’t cover it. Every step is in the log, so a mistake can be traced and the rule behind it fixed.
Does it handle Hebrew properly?
Current models read and write Hebrew well, including text that mixes Hebrew and English, and Russian too. There are weak spots: poor scans, handwriting, slang and very specific professional jargon. That’s why we test on your real documents and messages before launch, and why drafts meant for customers go through approval until you’re happy with the tone.
Will it work with the tools we already use?
Usually, yes. If a system has an API, the agent reads and writes through it; Gmail, Google Sheets, most CRMs and the WhatsApp Business API all have one. If a system can only export files, the agent works from exports or a shared folder. If there’s no sensible way in, we’ll tell you at the start rather than halfway through the project.
Where does our data go, and who owns the agent?
Your data stays in your systems; the agent reaches it with keys you grant and can revoke. The text it processes goes to the model provider through its business API, and we review that provider’s terms on data retention and training with you. We send only what the task needs. Ownership of the code and the accounts is agreed in writing before we start.
How much does an AI agent cost, and how long until it works?
It depends on the task: how many steps and systems are involved, how much the agent writes rather than just reads, and how many items it handles. There’s a one-time build cost and a running cost per use. Starting with one narrow task keeps both under control. After a short call, you get a written estimate of cost and timeline.
Got a task that eats up your team’s week?
Tell us about it on WhatsApp or in the short form. We’ll ask a few questions, tell you honestly whether an agent is the right tool, and if it is, suggest where to start.
Name one task your team dreads every week, and we’ll tell you honestly whether an agent can take it over.
Prefer to talk? Call050-854-2579