New AI agent puts a company’s own knowledge to work – at a predictable cost
For most businesses, AI still means a chat window: you type in a question and get an answer back. But that AI knows nothing about the organisation it is meant to be helping, and the real work is still left to people. Leading enterprise search provider searchit is now taking things a step further with the searchit AI Agent – one of the first agentic AI solutions to turn a company’s entire body of data into a memory for AI, at a price mid-sized firms can plan around. The agent can access file shares, mailboxes and internal systems, understands what is in them and acts on it: it compares contracts, drafts letters, produces Excel analyses and PDF reports, and sends them out when asked to. In short, the passive chatbot becomes a digital colleague.
What AI lacks isn’t intelligence – it’s memory
A language model can write, summarise and reason. What it cannot know is which contracts a company has with a customer, which invoices are still unpaid or what went into the last quote. “AI doesn’t need a better model – it needs a memory,” says Christoph Wendl, CEO of searchit. “And businesses already have one, in their documents, inboxes and shared drives. We open it up with searchit, and the agent puts it to work.”
To do this, searchit uses AI to read and summarise every document, picking out the people, companies, places and amounts mentioned. From this it builds a knowledge graph showing how customers, contracts, invoices and contacts relate to one another – even when the same company is spelt differently in two documents. More than 100 sources can be connected, from file servers, SharePoint and Teams to email, specialist databases and web sources.
A colleague that rolls up its sleeves
The searchit AI Agent does more than answer questions: it carries out tasks, with around 60 tools at its disposal. If someone in sales asks it to compare two framework agreements with customers, for instance, it finds the contracts via the knowledge graph, reads them in full and hands back an Excel spreadsheet setting out the term, scope, payment terms and discounts. If the accounts team needs a payment reminder, it goes through the list of outstanding invoices and earlier reminders and prepares a draft in Word, which is only sent once a person says so. And when asked to sum up a customer on a single page for the sales team, the agent gathers the contracts, outstanding invoices and contacts, produces a customer profile as a PDF and emails it over.
The agent can also sort mailboxes, book appointments, tag entire document collections, redact personal data and even operate web applications that have no API. Recurring jobs such as “show me the new tenders every Monday” run automatically on a schedule. Staff can reach it at their desks, in the searchit client and on the move via messaging apps such as Signal or WhatsApp.
A week’s work in a day
A customer project from 2026 shows how much manual work can be handed over. The head of sales at a construction company with around 200 staff was preparing a bid for a tender: reviewing the criteria and standards, going through previous quotes and service catalogues, pulling in clauses from the company’s terms and conditions and bringing subcontractors on board by email. This used to take around a week. With the searchit AI Agent, it took around one working day.
“The biggest cost is rarely the time spent searching,” explains Wendl. “It’s the work people still do by hand because no single system holds everything in one place: gathering documents from several systems, rebuilding the same report every week, copying figures out of a web application. That is exactly the work the agent takes on.”
The user’s permissions, not a master key
An agent that can change files and send emails needs clear limits. That is why the searchit AI Agent always works under the login of the person who gave it the task: it can only see and do what that person is allowed to see and do in the company’s systems. Each group of tools can be switched on or off by role, and every step is logged with the user, time and outcome. Password and PIN fields are removed before any text reaches the language model.
The data stays in an installation that belongs to the customer alone, and the customer decides where the agent runs: on staff PCs under Windows, Linux or macOS, centrally on a Windows terminal server, or in the cloud on a dedicated virtual machine. The EU AI Act is catered for too: every AI-generated field records the model, version, time and prompt ID, in machine-readable form and in exports.
Find out more about the searchit AI Agent:
