Blu OS on your existing computers
Blu Mesh joins suitable machines you already own into one AI system, working as one. The fastest way to test private AI on one workflow, with nothing new to buy.
Early access is open
Plug. Train. Relax.
Blu Box is your organization’s AI, on your premises. It runs Blu OS, which can also turn the computers you already own into one private AI system. Staff get agents that read, draft and answer, without sending files to a public cloud.
Designed and built by OceanBlu in Dhaka. NDA by default.
One system, two ways to run it
Start on the machines you have. Move to a Blu Box when you need more.
Blu OS on a Blu Box appliance
Blu OS preinstalled on proven AI hardware, sized to your team: Lite, Plus or Enterprise. Plug it into your network and we do the rest.
Blu Box tiers
The higher the tier, the more power and the tighter the seal.
Three tiers: Lite, Plus and Enterprise. Enterprise is fully air-gapped, our most protected tier.
Tier 1 of 3
Blu Box Lite
Small offices, shops and clinics (5 to 50 people)
- Hardware class
- Mac mini M4 Pro class
- Security
- Local-first, with optional cloud reasoning
Tier 2 of 3
Blu Box Plus
Companies, hospitals and banks (50 to 1,000 people)
- Hardware class
- Mac Studio or NVIDIA DGX Spark class
- Security
- On-premise only
Tier 3 of 3
Blu Box Enterprise
Government and defence
- Hardware class
- Rack or multi-node, solar option
- Security
- Fully air-gapped, no cloud path at all
The deployment problem
Today you get two bad options.
Public cloud AI
- Every file your staff send leaves your network.
- Costs rise with every token, every month.
- Models and prices change on the provider's schedule, not yours.
Build it yourself
- Which model? Which hardware? API or local?
- Agents hang, permissions drift, and nobody owns the fixes.
- You need DevOps and ML engineers you don't have.
Blu OS and Blu Box
- Models and knowledge run on hardware in your building.
- A fixed deployment and a support plan, not a meter that runs per token.
- We set it up, train it on your documents and keep it running.
We hit every one of these problems deploying our own first agents. Blu OS and Blu Box are what we wished we could buy.
Built for
For organizations that can’t paste their work into a chatbot.
Government & public bodies
Policy drafts, citizen records and internal files that must stay under your control.
Defence & security-adjacent teams
Work that can't touch a foreign server.
Enterprises & groups
Contracts, financials and HR files across several companies.
Universities & hospitals
Student and patient records, research data and admissions.
What's in the box
Hardware, an agent operating layer, and your knowledge.
New AI models come out every week. Blu OS isn’t tied to any one of them.
We swap in better models as they arrive; your agents and knowledge stay put.
The hardware
Your own suitable computers, or a Blu Box sized to your team: Lite (Apple Mac mini class), Plus (Mac Studio or NVIDIA DGX Spark class) or Enterprise (rack or multi-node, air-gapped).
Blu OS 1
Decides which model handles each task, manages agent memory and permissions, and keeps agents running. You won't have to think about any of it.
Your knowledge, ready to answer
Policies, contracts, manuals and past work stay on your network, so Blu and your agents answer from your documents, not from the internet.
Blu Agents and Blu Reef skills
Starting agents and sector skills for real estate, education, healthcare, legal and government workflows, adapted to your processes.
Inside Blu OS 1
The organization is the computer.
The Blu Focus family: the technology inside Blu OS that makes a private AI system fast, careful and easy to run.
Blu Focus
Gives the AI only the right context for each question, so answers are faster, sharper and cost less to run.
Blu Mesh
Joins the computers your office already owns into one AI system.
Blu Recall
Instant company memory: your policies, files and past work, ready the moment someone asks.
Blu Route
Uses the smallest model that can do the job, and a bigger one only when it's needed.
Blu Guard
Decides who, and which agent, may see what.
Blu Shift
The night shift: idle computers run research and reports after hours.
Blu Dock
Swap any open model in or out, like a driver.
Blu Console
The admin console: health, costs and the audit trail.
Ask Blu.
Blu is the assistant inside Blu OS. Staff ask it questions; it answers from your own knowledge, on your own hardware.
Blu Route picks the right model for each question behind the scenes, so staff just ask.
- AskQuestions about your policies, files and past work.
- DraftLetters, replies and reports in your house style.
- FindThe right document, in seconds, with only what you may see.
How your data moves
Local by default. You decide what, if anything, leaves.
Local first. Documents, questions and answers are processed on hardware inside your network.
Optional cloud reasoning, off by default. For very heavy tasks you can allow a call to a cloud model (Anthropic’s Claude). Before anything leaves, Blu OS masks sensitive details such as money amounts, email addresses and phone numbers, then restores them in the answer on your side.
Access you control. Blu Guard decides who, and which agent, may see what. Each agent only sees the folders and systems you grant it.
Your network
Staff ask, upload or forward work
Blu OS
Blu Route picks the model, Blu Guard applies permissions, Blu Focus brings the right context
Local models (Blu Dock)
Your knowledge (Blu Recall)
Answer, draft or routed file, back to staff
Off by defaultOptional: masked request to a cloud model, answer unmasked on your side
Use cases
Work your team does every day, handled on site.
Pick one to start. We measure it with you, and you see the results first.
Document search that answers
“What's the penalty clause in the Gulshan contract?” answered from your own files, with the source attached.
Drafting
First drafts of letters, tender responses, briefs and reports in your house style.
Intake and routing
Read incoming applications, forms or complaints; sort, summarise and send them to the right desk.
Internal help desk
HR, IT and policy questions answered from your handbooks.
Research briefs
Summaries of long reports and regulations for decision-makers.
Have a different workflow in mind?
Tell us about it
How a deployment runs
Plug. Train. Relax.
Step 1: Plug
We install Blu OS on your network, on your machines or a Blu Box, and connect it to the folders and systems you choose. No cloud account to set up.
Step 2: Train
We load your documents, set up one or two agents for your highest-value workflow, and your team tests them.
Step 3: Relax
Agents run daily. We monitor them, update models and handle support. Coming with Blu OS: Blu Console, a console for health, costs and the audit trail, and Blu Link, a support channel that never carries your data.
Early access
Be one of the first organizations to run Blu OS and Blu Box.
What you get
- Blu OS installed on your premises, on your hardware or a Blu Box
- One or two agents built for a workflow you choose
- Training for your team, and direct access to our engineers
- A written report of the results
What we ask
- A named owner on your side and access to the workflow's documents
- Honest feedback, weekly
- Permission to describe the results (anonymized if you prefer)
Tell us about your team and your first workflow. We’ll size it with you.
ConceptWho's building it
Built by a Bangladeshi team, overseen by people who understand sensitive work.
MD Al Asif Sikhon
Founder & CEO
Brig. Gen. Kazi Mohammad Kaiser Hossain, ndu, psc (retd)
Chairman
Hasibul Alam Siam
Head of AI & Automation
Is Blu Box available to buy today?
Blu Box is in early access. A small group of organizations get it first and help shape it. Join the early-access programme to be one of them.
Can it run on computers we already have?
Often, yes. Blu OS is designed to run on suitable machines you already own. We check your hardware first and tell you straight whether it's enough, or whether a Blu Box makes more sense.
Does any of our data go to the internet?
Not by default. Blu OS runs models and your knowledge locally. If you switch on optional cloud reasoning for heavy tasks, sensitive details are masked before anything is sent. You can leave it off permanently.
Which AI models does it run?
Open models that run well on local hardware, chosen for your workload. Blu OS is model-agnostic, so we update models as better ones are released.
What hardware is a Blu Box?
Proven off-the-shelf AI hardware running Blu OS, in three tiers. Blu Box Lite is Mac mini M4 Pro class, for small offices, shops and clinics. Blu Box Plus is Mac Studio or NVIDIA DGX Spark class, for companies, hospitals and banks. Blu Box Enterprise is rack or multi-node and fully air-gapped, for government and defence. We size it with you.
Do we need an IT team to run it?
No DevOps team is needed. We install, configure and maintain it. Your IT staff only need to give it a network connection and the access you approve.
How is this different from buying an NVIDIA DGX Spark ourselves?
A DGX Spark is a computer. Blu Box is the computer plus Blu OS, your indexed knowledge, sector templates, installation and support. You get working agents, not a box to configure.
Can it connect to our existing systems?
Yes. We start with the folders and systems your first workflow needs, and add more as you grow.
What does it cost?
It depends on your team, your hardware and your workflows. Talk to us and we'll size it with you. A fixed price and a support plan, never a per-token meter.
Who has the final say on what an agent sends?
Your people. Agents draft and recommend; your team approves anything that goes out. We set the approval steps for each workflow.
Your AI should work for you, not report to someone else's server.
Join the early-access programme. We'll set up a confidential briefing, under NDA if you need one.