AIEngineering studio · Israel
AI agents that work inside your systems. Not next to them.
Agent Hub Guru is an AI engineering studio in Israel, founded and run by Ofer Shahar. We design, build and operate AI agents and business automations connected to your CRM, WhatsApp, email and company knowledge. When an agent is not sure, it hands the case to a person.
- First discovery call is free
- Reply within 24 hours on business days
- Hebrew and English
Updated · Ofer Shahar, founder
What we build
- 01
AI strategy and process mapping
We sit with the people who do the work, break each process into steps, mark the repetitive manual ones, and decide together what is worth building first and with which tool. The output is a process map and a prioritised plan, not a slide deck.
Full page (Hebrew) - 02
Custom AI agents
Customer-service agents grounded in your approved knowledge, sales agents that qualify leads and hand them to a human rep with a three-line summary, and operations agents that work against your CRM, email and reports. Every agent has explicit escalation rules: what it does alone, and what goes straight to a person.
Full page (Hebrew) - 03
Business automation
Automations that connect forms, email, WhatsApp, CRM, Google Sheets, Notion and Monday, with AI added only where text has to be understood: summaries, classification, lead scores. Bad or missing data goes to a manual review queue instead of disappearing.
Full page (Hebrew) - 04
Knowledge bases and grounded chatbots
Chatbots that answer only from the documents, policies and catalogues you approved, cite the section they used, and say they are not sure — then open a ticket for a human — when no source fits.
Full page (Hebrew) - 05
Integrations and internal tools
CRM, ERP, websites, forms and contact centres connected through REST APIs and webhooks, data synced with validation before every write, and small internal admin screens that show what synced and what failed. Failing systems are retried, then escalated to an owner.
Full page (Hebrew) - 06
Sales and marketing AI
Sales calls transcribed (with consent), objections and agreed next steps extracted, a summary written into the CRM card and a follow-up draft prepared for the rep. Nothing is sent to a customer until a person has read and approved it.
Full page (Hebrew) - 07
Operations and internal knowledge
Meeting summaries turned into tasks, procedures written from conversations and existing documents, internal search in plain language, and manager reports you can ask for in Hebrew — always run with the permissions of the person asking, read-only by default.
Full page (Hebrew) - 08
Control and safety for AI agents
Permissions by role, monitoring, and a fixed test set of real conversations that every change must pass: each new version is compared with the previous one and goes live only after approval. A version that fails stays out; the old one keeps working.
Full page (Hebrew) - 09
Security-operations agents
Agents that take SIEM, firewall, EDR and cloud alerts, enrich them from logs, compare them with earlier alerts and propose a severity, so the security team starts from a ticket with full context. Support for ISO 27001, SOC 2 and GDPR work. The agent never blocks or changes policy — a person decides.
Full page (Hebrew)
Claude · ChatGPT · n8n · Make · WhatsApp · Supabase · Vercel · Monday · Google Sheets · Notion · REST APIs + webhooks
How an engagement works
- 01Discovery and mappingWe learn the process with the people who run it. Deliverable: a specification document.
- 02Strategy and tool choiceOnly tools that pay off — preferably ones you already use. You never pay for technology you do not need.
- 03Build, integrate, trainThe agent runs in a test environment before it meets a real customer, and your team is trained.
- 04Measure and improveWe check what the agent actually does and improve it monthly.
Engagement options
Prices are quoted per project, after the discovery call.
Basic
Mapping and a first agent
- Discovery session
- Mapping of 3–5 processes
- One basic agent
- Short team training
- Recommendations document
Growth most chosen
A custom agent connected to your systems
- Full process specification
- A custom agent
- 1–2 integrations
- A knowledge base
- Testing, training and initial support
Premium
Several agents, management and monitoring
- Agents per department
- Integrations across your systems
- Management and monitoring dashboard
- Complex end-to-end automations
- Monthly support and optimisation
Questions people ask before starting
Do we need technical knowledge to work with you?
No. We need you to know the work: how requests arrive, who handles them, where data lives and what gets stuck. We take the technical part — model choice, integrations, permissions and testing — and hand over short training and documentation at the end.
How long does an implementation take?
It depends on scope. One agent on one well-defined process is much shorter than a project that connects several departments and systems. After the discovery call you get a written, phased timeline before any work starts.
Is our data secure?
Agents get only the access they need, with minimal permissions, in accounts you control. Keys and passwords are never stored in code. Every agent runs in a test environment first and goes live only after your approval. For ISO 27001, SOC 2 or GDPR we build the documentation and controls accordingly.
Will AI replace my employees?
The goal is to remove repetitive work — typing the same answer for the hundredth time, copying data between systems, summarising meetings. Decisions, sensitive conversations and customers who need a person stay with people.
What happens when the agent does not know the answer?
It says it does not know and hands over to a person, with the full context of the conversation. Our agents answer only from information you approved; unanswered questions are collected and used to update the knowledge base.
Who owns the agent and the code?
It depends on the model. If you buy a full product, you own the code and the agent. If you use our existing agents as a service, the code stays with the studio and your data stays yours. This is written explicitly in the proposal.
Does it work in Hebrew and on WhatsApp?
Yes. The models we use, such as Claude and ChatGPT, read and write Hebrew well, and every agent is tested on real customer questions including typos and slang. Agents can answer on WhatsApp, on a website, by email or inside an internal system, and update a CRM, Monday or Google Sheets during the conversation.
What are the ongoing costs after launch?
Two kinds: usage, because model providers charge by the amount of text processed, and infrastructure — hosting and sometimes an automation tool or WhatsApp Business API subscription. The proposal includes an estimate for your volume, with a usage cap and an alert. Monthly support is optional.
Describe one process that eats your team's time.
First discovery call is free. Reply within 24 hours on business days. Hebrew or English.