Estate agency / In-house expertise & website chat
Your firm’s expertise. An assistant that works with it.
For an estate agency, AITJE linked Realworks, email inboxes and Google Drive to its own AI assistant. Staff chat with the knowledge relevant to their role. The website chatbot uses a separate, carefully selected knowledge base from the same environment.
- Updating knowledge
- 3× per day
- Internal chat
- GPT-OSS-120B
- Website chat · response time
- Approx. 2 sec.
- Website chat · current usage
- €2–4 / day
The response time and daily costs are the practical estimates provided based on current usage; they vary depending on demand and traffic.
The estate agent’s enquiry
The expertise, though.
Not the research.
At an estate agency, information is rarely stored in one place. Property details are in Realworks, appointments and correspondence are in the email inbox, and documents are in Google Drive. Finding an answer often means switching between systems and deciding for yourself which information is still up to date.
AITJE installed the Assistant on a Linux server, linked to an BOSGAME M5. The existing sources are consolidated into a dedicated knowledge base. Staff ask their questions in a single chat; the search layer retrieves relevant information, with the correct date, category and access rights.
Realworks
Property details and information from the existing estate agent platform.
Mailboxes
Correspondence and appointments as searchable business knowledge.
Google Drive
Documents and files from the linked business environment.
The existing systems remain the source. The assistant makes their knowledge accessible.
From source to usable answer
Connecting expertise.
Retain the context.
A bespoke ChatGPT-style environment, connected to the agency’s datasets. The difference lies in the preparation: the right knowledge must be in the right place before a model can work with it.
Synchronise
The knowledge from the linked sources is synchronised three times a day. This ensures that new and amended data is incorporated into the knowledge base.
A price has a date
What was true back then?
So what applies now?
A property may have a different asking price in February than in March. Simply finding text that is comparable in terms of content is therefore not enough. AITJE expanded its knowledge management system to include date, category and priority, so that the system can distinguish between current information and previous versions.
For a current query, the most recent information is taken into account. For a historical query, the previous situation remains searchable. Priority helps to rank relevant sources; access rights determine which knowledge is available in the first place.
Schematic example of a home
Findable for a historical query
Relevant for an enquiry about the current price
Source + date + category + priority + access rights
The internal environment
Proprietary models.
In-house knowledge base.
The internal assistant uses GPT-OSS-120B on the BOSGAME M5. The vector database is also stored there. The Linux server and BOSGAME together form the in-house environment in which the Assistant and the links are configured.
GPT-OSS-120B · answers
The open-weight chat model formulates responses using the knowledge retrieved by RAG. No external API token account is created for local model processing.
OpenAI · model information ↗Harrier-OSS-V1-27B · embeddings
Microsoft’s embedding model makes the content semantically searchable, enabling the search layer to search by meaning. Microsoft reports a score of 74.3 on Multilingual MTEB v2 and top performance at launch. This is a model benchmark, not an accuracy percentage for this knowledge base.
Microsoft · model map ↗Memory and model selection
The M5 ‘BOSGAME’ from the product link features a Ryzen AI Max+ 395 and a Radeon 8060S, with configurations of 96 or 128 GB LPDDR5X. The CPU and GPU share the memory.
Harrier has 27 billion parameters. The weights alone on 16-bit theoretically require approximately 54 GB; execution and context also require memory. GPT-OSS-120B is also a large model. Indexing and chat are therefore separate components in the design. This description does not assume that both models are stored in memory simultaneously in their unquantised form.
BOSGAME M5 · specifications ↗Knowledge tailored to your role
A single environment.
Targeted access.
Employees access the assistant via a dedicated URL with an authentication code. The connection to the local environment is set up using VPN/ngrok and authentication. Network access and access rights complement one another: a connection alone does not grant access to all company data.
Categories are assigned to roles. For example, an employee can access property information without having access to the financial records. The knowledge manager determines which sources are available for searching and responding.
How the access profiles work
Staff member
An employee can work with the assigned property and process information. Confidential financial categories are not automatically available.
- Allocated housing data
- Working arrangements
- Shared documents
- Sensitive financial data
- Knowledge beyond one’s own role
Illustration of the distribution; the specific roles have been defined for the agency.
From internal success to the website
The same basis.
A different target audience.
Following its implementation internally, the knowledge base also became the foundation for a chatbot on the website. This is the ‘AITJE Assistent’ in a different environment, adapted for visitors via AITJE Custom. The website information is supplemented with selected, approved knowledge from the internal environment.
The website chat has its own vector database. The server managing this has an authorised IP address for the synchronisation connection with the BOSGAME. Only the selected categories are transferred. This provides the public chatbot with useful knowledge without access to the full internal environment.
Internal knowledge base
Realworks · mailboxes · Google Drive
In-house website knowledge base
Website content + selected knowledge
GPT-6 Luna makes the query suitable for searching.
The website’s own knowledge base provides excerpts with source and date.
The model formulates a response based on the data it has found.
Website chat in practice
Quick reply.
Traceable leads.
In this application, the chatbot responds in around two seconds and generates traceable leads.
- Current daily running costs
- €2–4
- Response time in this application
- ≈ 2 sec.
A light model with specific context
For the website, AITJE chose GPT-6 Luna: an affordable external model that responds using the selected knowledge. The standard rates for short context are $0.10 per million input tokens and $0.50 per million output tokens. The €2–4 per day is the stated average based on current usage, not a fixed price for the entire installation.
OpenAI · Rates checked on 4 October 2026 ↗A fallback with another provider
In the event of problems with a model API, the workflow can switch to other models and a different provider, including Mistral Small 4 from the French company Mistral. The selected knowledge base remains the basis for the response, even when the model route changes.
Mistral · Small 4 ↗What ‘AITJE’ delivered
Knowledge that works for you.
A personal assistant for staff, a knowledge base with date, priority and role-based access, three synchronisations per day and a separate website chat with selected knowledge. The models support the process; the integrations and knowledge management make the solution usable.
This case study demonstrates how ‘AITJE Assistent’ can be deployed in a hybrid manner. Internally, model processing remains under the agency’s own management. For the public, the chatbot uses a cost-effective external model. ‘AITJE Custom’ connects the sources, sets up the synchronisations and adapts the application to the agency’s work.
Other case studies
Development agency / Assistant, Coder & model management
Ten models. One team. Control and independence.
AITJE set up a shared AI environment for a development agency, featuring Assistent, Coder and ten models to choose from. A shared token pool of €350 per month replaces nine subscriptions to €200. Model usage and token costs per task can be tracked via Asana.
View the approach
Marketing / customer management, tickets & business knowledge
350 clients. One central workspace.
Following a takeover, a marketing company’s client base grew from 150 to 350. AITJE brought together client data, tickets, projects, invoicing information and call transcripts in a central hub. Staff can search for and chat with their business experts, with access based on their role and AI integrated into an existing Mac mini.
View the approach
Marketing agency / level design & in-house coding agency
25 New game levels every week. With no external token costs.
A marketing agency took over an existing mobile puzzle game but ran into difficulties when creating new levels. AITJE analysed the game, built a reusable level design system and linked AITJE Coder to Qwen on its own server. Now, the agency 25 creates new levels every week from a single assignment in OpenCode.
View the approachMOVING FORWARD TOGETHER WITH AI
Your next step begins with a question.
Got an idea, facing a challenge, or simply curious? Tell us what you want to achieve. AITJE will work with you to find a solution.
