Better AI Context Management With Privacy. Case Studies: Plurality, Ekai.

Earlier this year, I had discussed at length about AI memory challenges and potential solutions. I mentioned a couple of names there who had been working on AI context management, as I firmly believe that is how we solve the memory constraints.

Here, I will further explore two projects that have progressed in this aspect and offer a glimpse of the AI future that could take our experiences to the next level.

Plurality & portable AI context

Consider this: what if you did not have to start over constantly, or to reiterate and summarize previous conversations whenever you switch models or start a new session?

Plurality started by trying to answer this. Their aim: to develop an Open Context Layer. As a result, you could autonomously store your data and chats in specific memory buckets. This would not only help with better context management, but also enable sharing the context as needed.

AI Context Flow

Billed as a fix for the problem of portable AI memory, Plurality built AI Context Flow. The tool works with ChatGPT, Claude, Gemini, Grok, and Perplexity.

One of the USPs of AI Context Flow is the ingrained privacy control. It thus differs from most other solutions, where cloud storage and data mining are part of the deal, mostly without the knowledge or consent of the users. Plurality teamed up with Oasis for a TEE-based, privacy-first context management, where you can decide what AI can see, you can revoke access at your discretion, and have complete data ownership.

Let’s now see how Plurality and Oasis address the privacy question.

Privacy problem

AI Context Flow solves one problem but opens up a new challenge — the infrastructure issue. Helping portable context flow through other servers is tricky and poses inherent trust gaps.

The easiest answer that springs to mind is adopting the Model Context Protocol (MCP). It enables agents to access tools, data, memory, and APIs. The infrastructure is powerful and growing as more and more AI products get built on top of it. But the privacy pain point lingers.

Most people, when they talk about AI privacy, are thinking about the model. But the underlying infrastructure is where most privacy is exposed. When a tool uses MCP, the MCP operator can see everything running through the server. So, you get the portable context but without privacy guardrails. Plurality faced this issue while building AI Context Flow, where user intent, history, and sensitive information were vulnerable to third-party access.

Oasis solution

Oasis comes into the picture by offering its Runtime Off-chain Logic (ROFL) for confidential computation to fix the infrastructure exposure dilemma. Now, the context can flow through the MCP server running inside a Trusted Execution Environment (TEE). This means:

• The operator cannot read context in memory
• The host OS cannot tamper with execution
• Code running is provable with remote attestation

So, when Plurality runs its MCP server on ROFL, the confidentiality of the infrastructure layer is not rooted in mere trust assumptions but actually verifiable. Context is stored and processed within the secure enclave away from prying eyes, and visible and accessible only to the model as needed.

Plurality next

Privacy-focused context management opens up new opportunities. In the next phase, Plurality is focusing on building a context marketplace on top of this foundation.

Simply stated, the context marketplace aggregates the information and knowledge that usually stays fragmented and in silos to be available in packages that anyone can opt to inject into any AI tool. Here, too, the privacy constraint is the potential deal-breaker. And Oasis, with its privacy layer, enables sharing and monetization of context without accessibility to the parties facilitating the transaction in the marketplace.

Ekai & AI context layer

The trade-off of using AI extensively is our dependency on the context that we have built meticulously over time, fine-tuning prompts and accumulating layers of responses.

Ekai has built a developer-focused solution that can seamlessly and cost-efficiently switch among various AI models via smart model routing without losing precious context. Oasis provides the trustless and verifiable privacy engine that such a context engine needs to function.

It addresses two major pain points in AI today.

• Access control: Access sharing is a common practice among users. They let friends and colleagues share access to AI models for ease of use. This is most prevalent while collaborating on projects. Divulging your actual credentials is a high-risk proposition.
• Agent context: Sharing access is easy; sharing context is the real challenge. Say, you are working with an agent for an hour, and it has accumulated patterns, decisions, contents, outputs, and the logic connecting everything. This context lives in plaintext and is shareable. But whenever there is a new session, the context is gone, or it is a nightmare to locate and reference it. Sometimes, even within a single session, the context degrades as the session timing progresses, leading to hallucinations. Summarization works with limitations, as compressed details tend to get lost in the AI’s memory architecture.

Ekai is trying to solve this by becoming the context engine for AI agents.

Contexto

Designed as the context layer for AI agents, Contexto currently works with OpenAI, Anthropic, Google, xAI, OpenRouter, and Groq, while the gateway runs on ROFL. Since it is built as an OpenClaw plugin, migration is not necessary.

From an architectural point of view, Contexto works by keeping the main thread clean by storing and indexing older work. So, when the agent calls for it, the tool retrieves the relevant context automatically.

All tasks and subtasks run in isolation and are handled by scoped sub-agents who will return contextualized outputs and assembled context in short and simple results. However, nothing gets thrown out, and if you need to trace the full details, they are recoverable on demand.

What Oasis adds

Ekai’s Control Plane enables storing encrypted API keys on-chain via Oasis Sapphire, so that you can delegate access with granular controls such as model restrictions, spending limits, and instant revocation. As decryption takes place only within the secure enclave or the TEE, it allows access sharing without credential sharing.

Now, context. As our session windows with AI models grow, the more we have to lose if trust fails, since the context plaintext is recorded for anyone with access to see or manipulate. This context confidentiality issue is solved with ROFL running the context layer inside a TEE. Result: context is stored, indexed, and retrieved within the secure enclave without ever exposing the plaintext.

Ekai next

As of now, both the Control Plane and the context layer as an OpenClaw plugin are live on mainnet. The next phase involves making persistent context private by default as they design the bringing of the context store into the ROFL enclave.

This would set the stage for agent-to-agent context routing. When this becomes possible, your agent will be able to collaborate with agents belonging to anyone. These disparate agents can then share relevant context with baked-in rules and privacy policies.

Final words

I am reminded of Oasis AI head, Marko Stokić’s discussion about AI memory in his Forbes article.

We need confidentially computed, and verifiable portable AI context where the user can decide if and when they want to move across AI models, and the interoperable memory layer is subject to the user’s data sovereignty and privacy. Both Plurality and Ekai, adopting a privacy-first approach with Oasis’s tech stack, demonstrate their commitment to that future.

What is your take on these projects working to solve AI’s persistent memory problem? Let’s hit the comments section.
And if you would want to explore more about how Oasis can help you deploy verifiable agents, the journey starts here.

Originally published at https://dev.to on June 24, 2026.

Better AI Context Management With Privacy. Case Studies: Plurality, Ekai. was originally published in Coinmonks on Medium, where people are continuing the conversation by highlighting and responding to this story.