The BlueNexus team are constantly researching emerging trends within the AI sector. Earlier this week we came across an extremely interesting article which proposed the notion of focusing strongly on context within LLM training methods. We find this particularly interesting as it strongly aligns with our product offering and wider vision of how AI not only should be developed, how it must be developed.
What if the secret to building smarter AI agents wasn’t better models, but rather better memory & context? This is the core idea behind Yichao Ji’s recent writeup, which details lessons from developing Manus, a production-grade AI system that ditched traditional model training in favour of something far more agile - "context engineering".
From Training to Thinking
Rather than teaching an LLM what to think through intensive fine-tuning, Manus has been focusing on designing how it thinks, via structured, persistent, runtime context.
Key tactics include:
- KV-cache optimization to reduce latency and cost
- External memory layers that store files and tasks without bloating prompts
- Contextual “recitation”, for example agents reminding themselves of their to-do list
- Error preservation as a learning loop
- Tool masking over tool removal, to retain compatibility and stability
This approach points to a deeper shift in the LLM training debate, shifting from “prompt engineering” to context architecture, and it’s changing how intelligent systems are being built.
Diving Deeper
Ji’s article observes that developers still default to the “If the model isn’t good enough, retrain it" approach. But Manus demonstrates that this isn't scalable. It’s expensive, brittle, & hard to maintain across use cases. Instead, they show that by designing the right context window with memory, goals, state, & constraints, developer you can achieve robust agentic behavior 𝐟𝐫𝐨𝐦 𝐞𝐱𝐢𝐬𝐭𝐢𝐧𝐠 𝐋𝐋𝐌𝐬.
We don't necessarily see this as a "work around" but rather the new standard emerging, which is fantastic within the R&D lens of LLM training.
Obligatory mention that we carry some level of bias here, as this new standard plays straight into our wheelhouse.
Alas, BlueNexus Agrees
We wont sit here and "Shill" this approach from the roof tops, it’s fair to say this emerging standard aligns strongly with what we have been building.
The future of AI isn’t just about inference, speed or model accuracy, in our opinion it’s about relevance, continuity, portability and coordination.
By this we mean:
- Knowing what data should (and shouldn’t) be in scope within any given prompt or automation
- Remembering past actions across sessions & various tools / 3rd party applications
- Structuring memory & state for reasoning, not just retrieval
As always, were interested in what other AI builders think?
- Are we overvaluing model complexity & undervaluing memory infrastructure?
- What makes context trustworthy, especially across tools, users, & time?
- Could context-based architectures unlock broader access to AI, without the cost of custom training?
- Is “context as code” the new OS for agents?
We would love to get a collective thoughts across the spectrum from anyone participating in this space. Feel free to add your colour to the conversation & start a dialogue with likeminded people in the comments below.
On this page
Similar Articles
Introducing the Universal MCP Server
February 10, 2026
The Context Problem in Personal AI
I've been building AI agents for personal productivity, and I kept hitting the same wall: getting my agent to access all my data in a way it could actually understand. The real challenge wasn't just connectivity - it was making that data useful to the AI while keeping it secure.
After wrestling with custom integrations, token management, and context window limitations, I realized we needed a fundamentally different approach. That's why we built the Universal MCP Server - a single endpoint that intelligently manages the bridge between your private data and any AI model.
What is the Universal MCP Server?
The Universal MCP Server is a remote Model Context Protocol (MCP) server that generates the optimal context window for any user prompt. Think of it as an intelligent middleware layer that sits between your data sources and AI applications.
Here's the core workflow:
- Prompt Analysis → The system receives a natural language request
- Source Selection → It identifies which data sources contain relevant information
- Intelligent Retrieval → Pulls data via third party APIs, MCP servers, Databases and more.
- Context Synthesis → Compresses and formats the most relevant information
- Structured Response → Returns optimized JSON or text to the LLM
But here's what makes it different: instead of dumping all available data into the LLM's context window, it acts as a Context Engine that filters and optimizes information before it reaches the model, significantly increasing performance and accuracy.
The Architecture: Two Layers Working Together
The Context Engine (Intelligence Layer)
When you ask something like "What did my team discuss about the Q4 budget?", the Context Engine doesn't just search for keywords. It:
- Locates the Q4 budget document and identifies recent comments
- Searches your company Notion for meeting notes
- Pulls relevant meeting transcripts from AI note taker (ie: Fireflies)
- Compiles all this data into a coherent narrative
This isn't simple aggregation - it's intelligent context formation. The engine understands relationships between different data types and prioritizes information based on relevance to your specific query.
The Universal Bridge (Connectivity Layer)
The second layer provides universal compatibility across AI platforms. Using the Model Context Protocol, it creates a single bridge connecting your private data to ChatGPT, Claude, Gemini, your own agents or applications. Basically you can connect to any MCP-supporting applications or code.
BlueNexus supports dynamic OAuth connectivity, so in many instances you can simply add the BlueNexus endpoint to your application:
https://api.bluenexus.ai/mcp
For some older clients, you will need to manually configure with a BlueNexus personal access token:
- Create a BlueNexus account
- Obtain your unique personal access token via the BlueNexus dashboard
- Use our one-line connection scripts to sync with any AI application
Why Current MCP Implementations Fall Short
Working with MCP servers extensively, I've identified three critical issues:
1. Tool Proliferation
MCP servers expose lists of tools that consume valuable context window space. Connect too many servers, and you've got hundreds of tools cluttering the LLM's context, making it harder for the model to understand what to call.
2. Context Generation Cost
Here's a fundamental truth about AI: what fuel is to cars, tokens are to AI. Every token consumed costs money and compute power. Current MCP implementations are economically suboptimal because they waste context window space on tool definitions rather than actual work.
You wouldn't drive your car to five different locations looking for the right wedding suit - you'd research and map out your purchase decision before getting in the car. Similarly, we shouldn't be loading hundreds of tools into an LLM's context window just to find the right one. For businesses watching API costs and eco-conscious developers concerned about compute power, this inefficiency is unacceptable.
2. Single-Tenant Inefficiency
Most MCP servers (excluding remote MCP servers) run on a per-user basis, which is incredibly inefficient, requiring a MCP server per user. We need multi-tenant servers that can support multiple users while still protecting individual tokens and data in a highly secure environment.
3. Credential Complexity
The current credential management nightmare is holding back AI adoption. Users face:
- Zero reusability - You connect your Google account to ChatGPT, then do it again for Claude, then again for your custom agent
- Repetitive authentication - The same OAuth dance, over and over, for every new AI app you try
- Developer overhead - Many MCP servers require you to register your own application, manage API keys, and handle OAuth flows yourself
This isn't just inconvenient - it's a fundamental barrier to AI becoming truly personal. Although dynamic client registration in the MCP spec will help, it doesn't solve the core problem of fragmented credential management across the AI ecosystem.
Our Solution: Unified, Secure, Intelligent
The Universal MCP Server addresses each of these problems:
Unified OAuth Management
This is the antidote to credential complexity.
Connect once, use everywhere - that's the promise of BlueNexus.
When you connect your Google account through BlueNexus, that connection becomes available across every MCP-enabled app you want to use. No more repetitive OAuth flows, no more managing dozens of app registrations. Your access tokens are stored in an encrypted database and injected in real-time when accessing third-party services, all within Trusted Execution Environments (TEEs).
Think of it as creating a digital AI brain that you can take with you anywhere. You don't need to register your own applications or run your own MCP servers - BlueNexus handles all the infrastructure complexity.
This means:
- Connect your accounts once, reuse them infinitely
- No application registration headaches
- No server management overhead
- Instant portability across AI platforms
Intelligent Tool Consolidation
By separating tool-calling logic from the LLM's context, we maximize the space available for actual work.
This is a fundamental optimization that delivers:
- Reduced costs - Fewer tokens means lower API bills
- Increased context capacity - More room for your actual data and conversation history
- Drastically improved performance - LLMs work better when they're not drowning in tool definitions
BlueNexus introduces cost and performance optimizations that a traditional LLM simply can't achieve on its own. Instead of exposing hundreds of individual tools, we provide a single, intelligent interface that routes requests appropriately. The Context Engine determines what's needed and fetches it - no tool spam in your context window.
Multi-Tenant Architecture with Privacy
Our server supports multiple users efficiently while maintaining complete data isolation. Each request carries a BlueNexus access token with user-specific scope, ensuring your data remains yours alone.
The Privacy-First Approach
I've always been passionate about data privacy and security, and I believe protecting user data isn't optional - it's fundamental. That's why we've built privacy into the architecture from day one:
- TEE-Protected Processing: All data handling occurs within Trusted Execution Environments
- Encrypted Token Storage: Access credentials are encrypted at rest and in transit
- Zero Knowledge Architecture: We process your data without storing or viewing it
This isn't just about compliance - it's about giving users confidence that their data isn't being consumed by big tech companies or accessed by others. While local processing is possible for technical users, we want a solution viable for everyone, which means providing confidential compute for AI infrastructure.
Real-World Applications
Health Intelligence
Connect all your wearable data and use AI to analyze your health patterns, provide personalized recommendations, and support your health journey. The Context Engine can pull from multiple sources - fitness trackers, health apps, medical records - to generate meaningful dashboards showing key health information in one place.
Productivity Workflows
The system excels at complex, multi-step tasks that typically fail with standard LLM setups. Meeting scheduling, for example, becomes a seamless four-step optimized process:
- Finding relevant documents
- Extracting participant information
- Checking calendar availability
- Sending invitations
Without the Context Engine, these workflows often fail due to tool-call errors, rate limits, and inability to manage complex logic. With it, they complete reliably and efficiently.
Financial Intelligence
Imagine asking "How much have I spent on electricity this year?" and getting an instant, accurate answer.
BlueNexus searches invoices across Gmail, Google Drive, Documents extracting payment totals, and returns a 12-month breakdown with citations. Or consider tax preparation - the system can aggregate receipts, categorize expenses, and compile documentation from across all your financial platforms.
The versatility of BlueNexus extends to any domain where context matters.
For end users, it means portable onboarding - use every app for the first time like you've used it forever. Your preferences, history, and context travel with you.
For app developers, it means context-rich awareness of your users from day one. Better engagement, better outcomes, and more conversions - because sales is always easier when you truly understand your customer.
The Technical Edge: Intelligent Context Model
Our flexible context model adds a middle layer of agentic capabilities that can analyze user requests and intelligently locate the most relevant data. It's not just about retrieval - it's about:
- Prioritization: Understanding which information matters most for the specific query
- Compression: Removing redundant or irrelevant data
- Formatting: Structuring information in ways LLMs can best utilize
- Relationship Mapping: Understanding connections between disparate data sources
This combination of external data connectivity, RAG systems, hybrid search, vector databases, and user memory provides a unified, powerful intelligence context engine.
Performance Expectations
While we're still gathering comprehensive metrics from production deployments, the architecture is designed to deliver:
- Significant token reduction by sending only relevant, compressed context
- Increased reliability through intelligent routing and error handling
- Faster response times by eliminating unnecessary data processing
- Higher quality results through better context formation
Getting Started
We're currently onboarding early users to the Universal MCP Server. The process is straightforward:
- Sign up for a BlueNexus account
- Connect your data sources through our OAuth flow
- Integrate with your preferred AI platform using our connection scripts
For developers, we provide simple copy-and-paste code snippets for connecting to existing AI agents. For consumers, we offer step-by-step guides for popular platforms like ChatGPT and Claude.
Final Thoughts
The future of personal AI depends on solving the context problem - getting the right information to AI models in the right format at the right time. The Universal MCP Server represents our approach to this challenge: a privacy-first, intelligent bridge between your data and AI capabilities.
By handling the complexity of data access, credential management, and context optimization, we're removing the barriers that prevent AI from becoming truly useful for personal productivity. The goal isn't just to connect AI to your data - it's to make that connection intelligent, secure, and effortless.
The Universal MCP Server is more than infrastructure; it's the foundation for a new generation of AI applications that can actually understand and work with your personal context. And we're just getting started.
Chris Were - BlueNexus Founder & CEO
06/02/2026
The Sovereign AI Shift Isn’t Coming, It’s Already Here
November 5, 2025
The ongoing discussion around “sovereign AI” sounds like a future-facing ideal rather than a current reality. local infrastructure, self-governed data, models trained on your terms are all pre cursers to achieving true "AI sovereignty". But recent initiatives across the AI sector potentially indicate that this "ideal" it's no longer a vision - it's happening.
It’s not just about national-scale deployments or GPU stockpiles, like the recent NIVIDA / South Korean alliance announced at the APAC summit. Sovereign AI is being built quietly inside enterprises, startups, and developer ecosystems, anywhere organizations want control over:
- Where their models run
- What data is used to train them
- How they comply with local laws
- Who has access to the outputs (and the logs)
This sets a clear mandate that as AI moves from novelty to necessity, the cloud-by-default mindset is starting to show its cracks. Companies are waking up to:
- Regulatory risk from black-box SaaS tools
- The fragility of building on closed APIs
- Ethical concerns around data reuse without consent
These factors are among a few examples of why we’re seeing an uptick in localized models, private compute clusters, and tooling built for “sovereignty by design.” Even small teams are asking: Can we keep our data in-region? Can we train on our own stack? Can we audit what happens under the hood?
This is a shift towards practicality where data governance is becoming a prerequisite to AI adoption, not just a bonus.
What This Means for Builders
If you’re building on today’s AI infrastructure, expect three trends to accelerate:
- Decentralized compute stacks: Not everyone needs to train a GPT-4. But many will want to fine-tune or host lightweight models on infrastructure they own or trust.
- Privacy-aligned design patterns: Users and enterprises alike are demanding revocable consent, encryption-at-rest, and zero data retention by default.
- Portable AI runtimes: The winning products won’t be locked into one cloud provider. They’ll work on-prem, on-device, or across federated environments.
At BlueNexus, We’re Betting on Sovereignty
From day one, we’ve believed that privacy shouldn't be a feature, rather a foundation to which any consumer facing AI product should build around. That’s why our architecture treats sovereignty as a default :
- Your context stays with you
- Your data is encrypted inside secure enclaves
- Your AI runs on infrastructure you control
As every aspect of our world from SaaS, to enterprise, to personal copilot usage, moves from dependency (on legacy systems) to autonomy (driven by agentic AI), we’re building the stack for people and teams who want to own their AI & the data its fed / produces, not just rent it.
Breaking the Walled Gardens: The Push for Tech Interoperability
October 28, 2025
Quick one today, as the team focuses on final checks and balances for an upcoming deployment. In this weeks article, we examine old business models coming to an end as new tech muscles its way through the "walled gargens" of big tech, to create a more open and interoperable future.
For years, Big Tech has profited by building "walled gardens", that is - closed platforms that tightly control apps, data, and user experience. These ecosystems, like Apple’s App Store or Meta’s social networks, maximize revenue and lock-in but restrict user freedom and innovation. Today, the tide is turning. Consumers, regulators, and even some companies are now demanding interoperability: seamless connection between apps, platforms, and devices. The result? A major shift in how digital infrastructure is being built.
The Signs of Change
- Cross-Platform Messaging: Apple will adopt RCS, improving messaging between iPhones and Androids. Under EU pressure, Meta is working on interoperable WhatsApp and Messenger platforms. The days of siloed chat apps are numbered.
- Smart Home Unification: Matter, an open smart home protocol supported by Apple, Google, Amazon, and others, now enables devices to work together regardless of brand.
- App Store Alternatives: In response to Europe’s Digital Markets Act (DMA), Apple is allowing third-party app stores and alternative payment systems on iOS—a sea change in how apps are distributed.
- Data Portability: New rules in the EU require platforms to let users move their data between services. Companies like Meta and Google now offer improved tools for exporting content and contacts.
- Open Social Networks: Meta’s Threads plans to integrate with ActivityPub, allowing users to interact across decentralized platforms. Tumblr and Flipboard are doing the same.
Consumer Preference
Interoperability matters to users because it offers:
- Freedom to switch platforms without losing data.
- Convenience of unified experiences across devices.
- Transparency and trust through open standards.
Surveys show over 80% of consumers prefer tech that plays well with others. In areas like smart homes and messaging, users are tired of being forced into one brand's system.
Why Companies Are Opening Up
- Regulation: Laws like the DMA in Europe are forcing change.
- User Expectations: Consumers demand flexibility and integration.
- Strategic Advantage: Interoperability can reduce churn and expand markets.
Even historically closed companies are making adjustments. Apple, for example, is embracing RCS and opening up iOS in Europe.
The Challenges Ahead
Interoperability isn’t without hurdles:
- Security and privacy risks increase when platforms open up.
- Loss of revenue from closed ecosystems (e.g., Apple’s App Store fees).
- Fragmentation if standards aren’t widely adopted or well managed.
The Bigger Picture
This is about more than messaging or app stores. It’s about redefining how digital systems are structured: from isolated silos to connected ecosystems. Users will gain freedom, developers will gain reach, and companies that embrace openness early may gain a competitive edge.
Final Thought
The push for interoperability is more than a trend it’s a structural shift. Platforms that prioritize openness, data portability, and integration are better positioned for the future. For users and builders alike, the walls are coming down—and the digital world is becoming a more open place to build.
