FAST LTA is developing its Silent AI enterprise search into Silent AI Pro, a local AI platform with projects, skills and agents on a 4U appliance. The company also presents a cost comparison with cloud subscriptions.
FAST LTA, a storage manufacturer based in Germany, is expanding its Silent AI product from an enterprise search tool into a broader on-premises AI work environment. The company presented the strategy on 8 October 2026 under the name Silent AI Pro. According to FAST LTA, the step responds to market demand that its current product covers only in part.
Three demand patterns
The company distinguishes three patterns of demand. The first is strong retrieval-augmented generation (RAG) with a simple chat interface, used to find and document knowledge; buyers are motivated by sovereignty and traceability. The second combines strong RAG with a full working environment, motivated by sovereignty and productive work in-house. The third requires no RAG but agents, skills and projects, driven by token costs and security. FAST LTA states that its current product fully serves only the first pattern. Silent AI Pro is meant to address the other two.
Platform and hardware
According to the company, the software keeps RAG, semantic search and source citation, and adds projects, skills, agents and artifacts. It runs on a new 4U hardware platform fitted with one to three GPUs depending on load. Six configurations, from R1000 to Z3000, share this platform. GPU capacity can be added during operation without migration or data relocation, the company says. The appliance is delivered preconfigured, operated in the customer’s own data center and managed by FAST LTA with support from Germany.
Operation without internet access
FAST LTA describes the appliance as operable without an internet connection: no open ports, no remote access and no telemetry. Users reach it through a browser inside the local network. Updates are applied only in scheduled maintenance windows. The company argues that a system not connected to the internet cannot be attacked from outside.
Local model via standard API
Applications such as ERP systems, ticketing tools, scripts or agents can address the appliance through an OpenAI-compatible API served by vLLM on the appliance’s own GPU, instead of calling a cloud model. Users can optionally select cloud models such as Claude or GPT from a drop-down menu. The appliance establishes that connection, and the default setting is local processing. According to FAST LTA, a request leaves the network only when a user chooses a cloud model.
Multi-tenancy and access rights
Several environments can run in parallel on one appliance, each with its own index, projects, chats, permissions, network and identity provider. A tenant can be a company, business unit, division or subsidiary. Only compute capacity is shared. Permissions are synchronized from AD/Entra ID and the source systems.
Cost model
The license is a one-time purchase, while service runs for a fixed term; the company offers a care contract of up to five years. FAST LTA’s cost calculator example assumes 500 users: 400 light users at 20 euros and 100 intensive users at 80 euros per month. That amounts to 192,000 euros per year in cloud subscriptions, or an average of 32 euros per user. Using an electricity price of 0.30 euros per kilowatt-hour and a three-year period, the company reports break-even after nine months for the R3000, with savings of 404,377 euros. For the H2000 it cites 16 months and 292,831 euros, for the Z3000 18 months and 255,547 euros.
Dr. Jakob Jung is Editor-in-Chief of Security Storage and Channel Germany. He has been working in IT journalism for more than 20 years. His career includes Computer Reseller News, Heise Resale, Informationweek, Techtarget (storage and data center) and ChannelBiz. He also freelances for numerous IT publications, including Computerwoche, Channelpartner, IT-Business, Storage-Insider and ZDnet. His main topics are channel, storage, security, data center, ERP and CRM.
Contact via Mail: jakob.jung@security-storage-und-channel-germany.de