Fundarem
Cosa fa Fundarem e come si comporta per la vostra azienda: i modelli di IA e dove ciascuno elabora una richiesta, dove sono conservati i vostri dati e cosa ne esce, come vengono approvate le modifiche, i passi dalla prima chiamata all’avvio, la conformità e i prezzi.
20 of 20 sections
Fundarem connects the software systems a business already uses into one map of the business, so AI agents, assistants and dashboards can work on it.
It is built for European businesses that want to use AI and keep to the EU rules. It is designed to run in the company's own Microsoft Azure, in an EU region, so the data stays in the company's own cloud.
The live demo shows it on a made-up company with made-up people. Nothing you do there touches a real system.
Sul sito:Live demo
Your systems keep running where they run today. Nothing is migrated or replaced: Fundarem reads from them, and writes changes back to them only where you allow it.
Each system is translated into one map of your business: objects such as customers, invoices, projects and people, the links between them, and the actions that can be taken on them. When the same customer is known in several systems, Fundarem finds the match by itself and puts it to your admin as a proposal; it counts on the map only after it is approved as your company's approval rules require.
Agents and assistants work on that map, never on a system directly. When a task needs an AI model, only the prompt for that one call goes to the model you chose, and the model keeps nothing between calls.
What is logged in your own cloud: every answered model call, with who or which agent made it, the model used and what it cost; every request an agent makes to your map; and every change, with who proposed it and who approved it.
Sul sito:How it works
Agents do routine work: they read, check and prepare, and every change they want to make is proposed; your admin decides which changes may be written back and which run automatically. Ready-made company agents come with the product and run on the model your company chose; your company switches one on through a proposal, and every run counts against your spending limit. Each person can also pick agents or create their own.
Every employee has their own assistant, opened from the corner of every page and bound to that person's own access. It reads the page the person is on, answers in plain words and follows the person's role, so a salesperson and the finance lead get different answers to the same question. It answers from all your connected data, explains and proposes; every change it proposes needs approval. By default it answers in text, and it builds a widget or a report when asked.
Each company has its own company brain: written entries on how your company works, such as how you price or who signs off a discount, each with an owner and a review date, kept in your own installation. The first-run setup proposes a starter brain from your answers, and a change to a shared entry needs the approval your company has chosen. People search every entry and propose new ones on the brain's page, and a shared entry can be limited to who may read it. The product's manual is inside your installation too, and the assistant answers questions about it with a link to the screen.
Dashboards are built from the same map, so every figure comes from the same joined-up data.
Your IT admin can add sections to the side menu beside the dashboard, for example one for people records: chosen kinds of records with chosen fields, from one system or several, shown to the roles the admin picks. A section reads the data each time it opens and never shows more than the person's role may read; personal records show a person their own record, while a manager and the IT admin see them all.
Sul sito:How it works
You choose the AI model that fits your rules, the quality your work needs and your budget. There are five kinds of offering: the default model, models processed only in the EU, models processed in any Azure region, frontier models, and a model on your own machine. Each is described below with every model in it, its licence and where it processes a prompt.
The frame is the set of offerings your company may use. It is chosen with us at setup, so a regulatory choice such as EU-only processing cannot be changed by accident later. An EU-only frame keeps every request processed in the EU.
Inside the frame, your IT admin picks which model takes which kind of task: a strong model for the hard work and a cheaper one for routine tasks. With EU-only processing, for example, the strongest model processed in the EU takes the hard work and a cheaper one processed in the EU the routine work.
We always offer the best models we can find and look for the lowest price, because we earn nothing on them. You pay for model use directly: we take no cut.
Sul sito:Models
The best open-weights models in the Microsoft Azure AI Foundry catalogue, one per vendor, deployed into your own Azure and billed on your Azure invoice, at open-model prices. The default model is one of them.
Your stored data stays in your EU region, but the prompt of each request is processed outside the EU: the default model is processed by a third party that serves it inside Azure, and Microsoft excludes that hosting from its EU Data Boundary commitments.
The models in this offering:
FW-Kimi-K3, from Moonshot AI. Licence: Kimi K3 License. It is processed outside the EU, by a third party serving it inside Azure; pay per use, on your Azure invoice.
FW-DeepSeek-V4.1-Flash, from DeepSeek. Licence: MIT. It is processed outside the EU, inside Azure; pay per use, on your Azure invoice.
FW-GLM-5.3, from Z.ai. Licence: GLM-5.3 License (MIT-style). It is processed outside the EU, inside Azure; pay per use, on your Azure invoice.
gpt-oss-120b, from OpenAI. Licence: Apache-2.0. It is processed outside the EU, inside Azure; pay per use, on your Azure invoice.
Sul sito:Models
For work where every prompt must stay in the EU. These are the best open-weights models sold by Microsoft with an EU Data Zone deployment, one per vendor: Microsoft processes each prompt inside Azure, only in EU member countries, under one Azure agreement.
The choice is smaller than the default's, and new releases reach the EU Data Zone later. Llama-family models are not offered.
The models in this offering:
DeepSeek-V4-Flash, from DeepSeek. Licence: MIT. It is processed only in the EU, by Microsoft; pay per use, on your Azure invoice.
Mistral-Large-3, from Mistral AI. Licence: Apache-2.0. It is processed only in the EU, by Microsoft; pay per use, on your Azure invoice.
Sul sito:Models
The strongest closed models sold in Azure AI Foundry, one per vendor, for the hardest tasks. They are deployed into your own Azure and billed on your Azure invoice.
Closed weights cannot be moved to your own machines. The prompt of each request may be processed in any Azure region worldwide, including outside the EU; for some of these models the model's vendor is the data processor. The anonymiser is on by default for them as for every model.
The models in this offering:
Claude Opus 5.5, from Anthropic. Licence: closed, under Anthropic's terms. It may be processed in any Azure region worldwide, including outside the EU; pay per use, on your Azure invoice.
gpt-6-astra, from OpenAI. Licence: closed, under OpenAI's terms. It may be processed in any Azure region worldwide, including outside the EU; pay per use, on your Azure invoice.
grok-4.7, from SpaceXAI. Licence: closed, under SpaceXAI's terms. It may be processed in any Azure region worldwide, including outside the EU; pay per use, on your Azure invoice.
Sul sito:Models
For full isolation: open model weights run on a GPU machine in your own Azure, in Norway East, Sweden Central or West Europe. No model vendor is involved, and no prompt leaves machines you run. Any other open-weights model whose licence allows it runs there too, including your own fine-tuned one.
You pay for the machine every hour it runs, used or not, and you look after it. The largest models need a multi-GPU machine, which costs far more; the models page shows the monthly range per machine.
The models in this offering:
DeepSeek V4.1 Flash (open weights), from DeepSeek. Licence: MIT. It runs on your own machine, in your Azure region; you pay for the machine by the hour.
GLM-5.3 (open weights), from Z.ai. Licence: GLM-5.3 License (MIT-style). It runs on your own machine, in your Azure region; you pay for the machine by the hour.
Kimi K3 (open weights), from Moonshot AI. Licence: Kimi K3 License. It runs on your own machine, in your Azure region; you pay for the machine by the hour.
Qwen3.8 2.4T A95B (open weights), from Qwen. Licence: Qwen3.8-Max License (MIT-style). It runs on your own machine, in your Azure region; you pay for the machine by the hour.
Sul sito:Models
A direct connection to the model vendor's own API, outside Azure, billed by that vendor. Your key to the vendor's API is stored in your own Azure subscription.
The vendor processes each prompt on its own service, outside Azure and outside the EU, as an independent data processor, so you need a DPIA and a transfer safeguard. The anonymiser is on by default for it as for every model.
The models in this offering:
Claude Opus 4.8, from Anthropic. Licence: closed, under Anthropic's terms. It is processed by the vendor on its own service, outside Azure and outside the EU; pay per use, billed by the vendor.
Sul sito:Models
The anonymiser is on by default for every model. Before a request goes to a model, it replaces with placeholders the values in the columns your admin marks for masking, and the names and patterned values it recognises, such as Norwegian national ID numbers, phone numbers and IBANs.
The answer comes back with the real values put back in, so people see real names while the model never did.
Your admin can switch it off per kind of personal data, one model offering at a time; every other kind stays masked. On a model on your own machine it can be switched off for that model altogether. What it cannot catch, such as nicknames, is listed on the compliance page.
Sul sito:How it worksCompliance
Fundarem is built so that everything it stores stays in your own Azure, in the EU region you choose at setup: the map of your business and its history, assistant conversations, the company brain and the log. Your database, files, credentials and model deployments are resources in your subscription, on your bill.
Nothing is stored with us. It is not a shared cloud service: if you part ways with us, there is nothing to hand back.
To bill per active user, your installation sends us one number each month, the count of active users, never who they are or any content.
Sul sito:How it worksCompliance
Your systems to your map: connectors read your systems into the map of your business on a schedule, inside your own Azure. In the first-run setup your admin picks which kinds of personal data are always masked before anything goes to a model, and tags each column of a new connector as it comes in.
Your map to the model: when a task needs an AI model, only the prompt and context for that one call leave, masked by the anonymiser, to the model offering you chose. Which offering processes it where is said on each model above. The model keeps nothing between calls.
The model back to you: the answer comes back into your installation, with masked values put back in. Nothing else leaves: not the map, not the history, not the log.
Every step is logged in your own Azure: every answered model call with the model used and what it cost, every request an agent makes to your map, and every change, with who proposed it and who approved it.
Sul sito:How it works
La scrittura nei vostri sistemi è una delle impostazioni che il vostro amministratore gestisce in Fundarem, accanto a ruoli, modelli e anonimizzatore. Resta spenta finché il vostro amministratore non la accende: prima di allora non viene scritto nulla in un sistema collegato.
Ogni connettore dichiara le operazioni di scrittura che può eseguire nel proprio sistema, come creare o aggiornare un record, e solo quelle possono essere accese. La libreria dei connettori mostra quali connettori possono scrivere.
Il vostro amministratore decide in tre passaggi: prima quali oggetti di un sistema possono essere scritti; poi quali operazioni di scrittura su quegli oggetti sono accese; poi, per ruolo e per persona, chi può usare ciascuna operazione e come viene decisa: due approvatori (la approva una persona diversa da chi l’ha chiesta), un approvatore, semiautomatico (si esegue da sola quando può essere annullata, altrimenti decide un approvatore) o automatico. Il vostro amministratore sceglie anche chi approva: i responsabili, oppure ruoli e persone scelti.
Una richiesta non va mai oltre ciò che la persona o l’agente che la fa può fare, e ogni richiesta, decisione e scrittura viene registrata con chi, cosa e quando. Se un sistema non risponde chiaramente se una scrittura è andata a buon fine, non viene inviata di nuovo: viene segnata come esito sconosciuto, e il vostro amministratore controlla il sistema prima che qualcuno riprovi.
Il vostro amministratore può cambiare tutto questo in qualsiasi momento, quando collega un sistema o dopo, in Admin → Orchestratore → Azioni per sistema. Una modifica viene proposta lì e vale non appena un responsabile l’ha decisa.
Sul sito:How it worksConnector library
A ready connector from the library: connectors are sorted by category, and each one says what it reads.
The system's own MCP server: a system that already offers one is connected as it is, and only the tools on its allow list, agreed with you at setup, are let through; your admin chooses the tools, fields and columns on a screen.
A custom connector: for a system with neither, we provide a connector and set it up with you.
All three end in the same map of your business, and before any data flows your admin confirms on a screen what a new connector brings in. After go-live your admin adds more connectors from the library at any time, and new ones reach the library with software updates. Missing one? Request a connector and we send a quote.
Your IT admin picks connectors and connects them, tags each column of a new connector for masking as it comes in, and chooses in the first-run setup which kinds of personal data are always masked.
A manager decides which connectors are switched on, and whether for everyone or for managers only. A change to a shared entry in the company brain needs the approval your company has chosen for the brain.
The sign-in details for your systems are entered by your IT admin and kept in your own Azure, not in Fundarem's own data.
Each person's assistant and agents see only what that person's role may see.
Sul sito:Connector library
1. Contact us: tell us about your business, the systems you use and the AI model you prefer. 2. The call and the contract: we talk through your systems and your compliance needs, and agree the contract. 3. Your onboarding profile: the systems to connect, the AI models you may use, your region and your plan.
4. One standard edition of the software for every company, delivered ready to set up. 5. You choose who sets it up in your Azure: yourselves, with our steps, or us for you. 6. Fundarem is installed in your Azure, with sign-in through your own company accounts.
7. Your IT admin sets your preferences in the first-run setup and enters the credentials for your systems. 8. We check the installation in your Azure before it goes live. 9. Live: your company starts working with Fundarem.
After go-live, when we publish a new version, your admin approves it in the admin panel or sets an automatic window for it, and the installation installs it by itself.
Sul sito:How it worksContact us
Fundarem runs in one dedicated part of your Azure subscription. Your IT admin gives us access to that part only, in the Azure portal, and you keep full ownership of your subscription.
After go-live, your admin chooses in the admin panel how much access we keep, and can switch it later: we stay in, to monitor the installation and assist, or limited access, only what is needed to run it. The admin panel records the choice, and your admin then makes the matching change in your own Azure. Fundarem then checks in your Azure that the change matches the level chosen.
You can take the access back at any time.
Sul sito:Contact usCompliance
On security: everything Fundarem stores stays in your own Azure, in the EU region you choose. The anonymiser, on by default, masks personal data before a request goes to a model. Every model call, request and change is logged in your own Azure. Our access is limited to one dedicated part of your subscription, which you can take back at any time.
The compliance page explains, in plain words, which EU rules apply when you use Fundarem, what Fundarem does about each, and links to the official text: the GDPR, the EU AI Act, the Schrems II limits on sending personal data out of the EU, and NIS2.
Norway and Iceland take the EU rules in through the EEA Agreement.
The compliance documents are published there too: how data flows, the data protection impact assessment, the data processing agreement, sub-processors, retention, your rights, the legal basis and the AI Act.
Sul sito:Compliance
To book a call, fill in the short form on the Contact us page, and we call you. You tell us about your business, and we talk through your systems and your compliance needs.
After the call you choose a plan: Self-serve (you set it up yourselves), Guided (we set up your dashboards and frontend with you) or Partner (specialists trained by us do it for you).
Before booking, you can try the live demo on a made-up company.
Sul sito:Contact usLive demo
Pricing is on request: tell us your systems and we send a quote. It is competitive with Microsoft Copilot.
You pay for the AI model directly: Microsoft bills it on your own Azure invoice, and a model vendor's own service bills you directly. We take no cut and look for the lowest price, because we earn nothing on the models. A self-hosted model needs its own machine, which costs the same each month whether it is used or not; the models page shows the prices.
A system that is not in the library is priced by how much work it takes to connect, with a one-time build fee and a monthly connector fee, both priced on request.
Spending on AI is always tracked, with a limit per request and per day that your admin sets on a screen of their own. Your admin sees, per month, the days each user was active and what each user's and each agent's AI use cost.
Sul sito:ModelsConnect your systems
Prezzo su richiesta: diteci quali sistemi usate e vi inviamo un preventivo. È competitivo rispetto a Microsoft Copilot.
Non c’è un costo di avvio.
Pagate il modello di IA direttamente: Microsoft lo addebita sulla vostra fattura Azure, e il servizio proprio di un fornitore di modelli vi addebita direttamente. Non tratteniamo nulla e cerchiamo il prezzo più basso, perché non guadagniamo nulla sui modelli. Un modello auto-ospitato ha bisogno di una macchina propria, che costa lo stesso ogni mese, che sia usata o no; la pagina dei modelli mostra i prezzi.
Progettata per funzionare nel vostro tenant Azure, in una regione UE; i dati personali vengono pseudonimizzati prima di andare a un modello esterno alla vostra installazione.
Il nostro accordo sul trattamento dei dati (GDPR art. 28). Ogni cliente lo firma con noi prima che inizi qualsiasi trattamento. Fundarem è progettata per funzionare nel vostro tenant Azure, in una regione UE.