Inferra by FetchSky
Your own AI. Inside your walls.
Inferra is a private LLM appliance for banks and financial institutions. The hardware and the top open models arrive together, running on your premises with no connection to the outside world.
- prompts leave your premises
- 0prompts leave your premises
- of prompts reviewable by compliance
- 100%of prompts reviewable by compliance
- top open models preinstalled, including Kimi, GLM, Qwen and DeepSeek
- 4+top open models preinstalled, including Kimi, GLM, Qwen and DeepSeek
- box to install, with users, guardrails and logs included
- 1box to install, with users, guardrails and logs included
In short
Inferra is a private, air-gapped LLM appliance for banks and financial institutions: purpose-built hardware with the top open models — Kimi, GLM, Qwen and DeepSeek — preinstalled and configured, plus user management, guardrails and compliance review of every prompt. It runs on your premises with no connection to the outside world, so every team gets a capable AI assistant and nothing leaves the building.
On-premise AI
Your data never leaves your vault.
Prompts, documents and answers stay on Inferra hardware inside your own data centre. Nothing travels to a cloud provider, nothing is used to train anyone else’s model, and compliance can review every prompt from day one.
Why can’t banks just use cloud AI?
Your teams want AI. Your regulator says no to the cloud.
Banks and financial institutions are often not allowed to send data to cloud AI tools. Meanwhile staff use them anyway, or go without. Either way the institution loses: either control, or the productivity every other industry is gaining.
What it is
What is Inferra?
A private AI assistant on a machine you own. Inferra is hardware with the best open models already installed and configured, plus user management, guardrails and compliance controls. Plug it in, connect your users, and every team has a capable AI assistant that never touches the internet.
Cloud AI tools
- Prompts travel to a third party
- Compliance sees nothing
- Models chosen for you
- Open to anyone with a login
- Needs an internet connection
Inferra
- Prompts stay on your premises
- Compliance can review every prompt
- Top open models, preconfigured
- Users, roles and guardrails you set
- Runs fully air-gapped
What’s inside
What’s inside the box?
Everything it takes, in one appliance.
| Layer | What you get |
|---|---|
| Hardware | A purpose-built appliance that runs the models on site |
| Models | The top open models, preconfigured and ready: Kimi, GLM, Qwen, DeepSeek and more |
| Users | User management with roles, so each team gets the access it needs |
| Guardrails | Policies on what the AI can and cannot do, set by you |
| Compliance | Review of every prompt, with an audit trail |
| Agents | AI agents that carry out multi-step tasks inside your environment |
Who uses it
Which teams use Inferra?
One appliance, every team.
Developers
Code in Visual Studio with an AI assistant, with no source code leaving the bank.
IT and DevOps
Draft scripts, troubleshoot incidents, create environments through agents.
Testing and QA
Generate test cases, test data and bug reports.
Risk
Summarise policies and exposure reports, draft analyses.
Operations
Process documents, draft responses, reconcile and explain exceptions.
Legal and compliance
Review and summarise agreements and regulations.
Marketing
Draft campaigns, product copy and customer communications.
Agents
What can Inferra’s agents do?
Ask once. The agent does the work. Inferra agents carry out multi-step tasks inside your environment. They do not just answer, they act, within the permissions you give them.
- You: Create a VM with a CI/CD pipeline for the payments team.
- Agent: Creates the virtual machine.
- Agent: Sets up the pipeline and connects the repository.
- Agent: Runs a first build and reports back.
Compliance
How does Inferra keep compliance in control?
Built so compliance can say yes.
Review
- Compliance can review every prompt
- Audit trail of who asked what, and when
- Reports for internal audit
Control
- Set users, roles and access per team
- Guardrails define what the AI may and may not do
- Agents act only within granted permissions
Contain
- Air-gapped, with no connection to the outside world
- Models and data stay on your hardware
- Nothing is used to train anyone else’s model
How it works
How does Inferra go from box to first prompt?
Step 1 of 4
Install
The appliance is delivered, racked and configured in your data centre.
- Install: The appliance is delivered, racked and configured in your data centre.
- Connect your people: Set up users, roles and guardrails with your IT and compliance teams.
- Switch on the teams: Developers, risk, operations, legal and marketing start using it.
- Extend with agents: Add agents for the repetitive IT and operations tasks that matter most to you.
Who sets Inferra up and keeps it running?
Engineers who stay. Inferra is delivered by FetchSky’s forward deployed engineers, who work next to your IT and compliance teams to install it, fit it to your workflows, and keep it running.
Read next
Frequently asked questions
[1]Does Inferra need an internet connection?
[2]Which models does Inferra run?
[3]Can compliance see what staff ask?
[4]Who can use it?
[5]Can it do more than answer questions?
[6]Can we use it with Visual Studio?
[7]Where does our data go?
Give your teams AI without giving up control
Inferra is opening to a small group of early institutions. Tell us about your environment and we will show you how it would run in your data centre.
