Dossier 04 · Banking & finance
A bank’s client memory, on its own servers
Everything a bank knows about a client is scattered across calls, products and years. Mnemosyne OS brings this context together in a single memory. The relationship manager finds everything in one place. Customer service has it in front of them instead of searching five tools.
This memory stays on the bank’s infrastructure, encrypted with AES-256 at rest when the bank turns it on, with a key the bank holds, and no third-party cloud. Banking secrecy, data residency and traceability require it.
Windows, macOS and Linux.
Local or your server
The model runs where the bank decides
Everything on the machine
The model runs inside the application (Qwen, Gemma, Phi in GGUF). Unplug the network cable: search and answers keep working.
Search local, answer in the cloud
Search runs on the machine. A cloud model writes the answer, with your own key or your Vertex project. The question and the retrieved passages go to that provider.
Your bank’s model
Your IT already runs a model? Connect it: any server that speaks the OpenAI-compatible API, at an address you set in the settings. The question and the passages go to that machine.
Compatible models
- Vertex AI
- Gemini
- OpenAI
- Anthropic
- Mistral
- DeepSeek
- MiniMax
- Qwen (local)
- Your server
A bank can keep everything on its own machines, with a local model or its own server. Hybrid mode is for public documents. For a pilot in your environment, talk to us.
Download Mnemosyne OS → Talk to us →Verified answer
Every sentence checked against its sources
Turn verification on, and a second pass re-reads the answer sentence by sentence. For each sentence, it looks for the passage that supports it in the documents served.
A quotation must appear word for word in the named document. Otherwise the sentence is flagged. The manager sees at a glance which sentences are supported, partial or unsourced.
The verdict stays attached to the conversation. You can export it with its evidence, and the cost of verification has its own line in the local journal.
At the bank A client’s context, in one place
The client file
Calls, letters, products, years of relationship. Mnemosyne OS reads these documents where they are. The relationship manager queries them from a single memory.
The manager’s question
The manager asks what was agreed at the last meeting. The answer cites the source document. A contract reference is found as written.
Check before answering
With verification turned on, each sentence of the answer is compared with the cited documents. The manager sees the ones no source supports.
Trace every call
Every call to a remote model is written to a local journal, with its tokens and its duration. With a local model or your own server, the question stays inside your network.
Questions
What a bank asks us before a pilot
- Where does client data live?
- On the bank’s machines. The engine reads the files in place, the index is local, and vault content can be encrypted with AES-256 at rest, under a key the bank holds. We host no documents.
- What leaves the machine?
- It depends on the model, and the bank chooses it. With a local model, documents and questions stay on the machine. With your own server, they stay inside your network. With a cloud model, the question and the useful excerpts go to that provider, under your key.
- Can an answer be checked?
- Yes. Every answer cites its documents. With verification turned on, a second pass compares each sentence with those documents and flags the ones none of them supports. Every call to a remote model is written to a local journal.
- How does a pilot start?
- With a conversation about your environment: which documents, which scope, which model. The pilot then runs on your machines, on the scope you choose.
We are looking for a bank or a fintech to run a pilot in its own environment.