This blog is co-authored with Zoe Weinberg and Matt Hawes at ex/ante, and is a follow-up to our first blog post on the topic, 'You don't own your memory.'
We need an open architecture that puts us in control of our memories while making their exploitation technically impossible. But how will this shift happen?
In order to discover possible implementations, we must understand how our data informs LLMs. The three predominant context engineering techniques are prompt design, retrieval-augmented generation (RAG), and fine-tuning—each of which inject context differently. The first two pull text into the prompt, while fine-tuning retrains the original model’s weights. Prompting helps guide but is vulnerable to edge cases. RAG dynamically cites references but shrinks the prompt window. Fine-tuning optimizes for a task but cannot be updated incrementally. The right systems won’t just personalize—they’ll do so modularly.
We need a rebel alliance: not a single product or company, but a diverse ecosystem of technologies that work together across the stack to demonstrate the tangible value of user-owned memory.
Most important to this alliance will be entrepreneurs who know how to build practical products that make data ownership matter to real users.
Some of these products will look like painkillers, solving urgent problems that users already feel acutely. Others will be vitamins, delivering 10x better experiences than today’s alternatives, or creating magical new products that deserve their own category.
Some will be overt champions of data ownership, while others will take a subtle 'trojan horse' approach, delivering data ownership and portability quietly in the background while focusing attention on delightful user experiences.
Users won’t adopt a new system just because it’s “the right thing.” They’ll adopt it because it works better. In the status quo marketplace of software, users usually must trade privacy for functionality. The opportunity now is to flip that equation — to create tools so useful that users don’t even realize they're benefiting from sovereign data.
Here are some key areas where we see opportunity:
Security: Foundational security infrastructure will pave the way for other experiences to emerge. Once data leaves a trusted environment, it can be infinitely replicated and misused. Rather than rely on contractual promises, we need systems that enforce privacy at the architectural level, making misuse technically impossible. Companies like Confident Security are building provably-private AI along these lines
Dev tooling: Instead of competing with ChatGPT's embedded memory, startups could collaborate to create a shared layer of user-owned data. Companies like Basic and Memory are building APIs and protocols reminiscent of Plaid's financial ecosystem, with authentication and database management serving as potential entry points.
Prosumer tools: Products like Highlight or Liminary drive productivity by connecting your workspaces (calendar, email, notes), so you no longer have to worry about context switching. Workshop Labs trains a private AI that thinks and works like you. Meanwhile, Sentience consolidates your life's data into a searchable, private memory bank that you fully own.
Consumer social: Consumer apps are proving a new breed of primitives becomes possible when users control their data. Patina helps you extract and generate artifacts from your camera roll. Shelf tracks your media consumption to showcase your taste. Fulcra makes sense of your wearable health data and daily habits so you can spend time more wisely. Meanwhile, decentralized social networks like Farcaster and Bluesky are pioneering user-controlled social graphs and portable identities, showing how social media can thrive without centralized data ownership.
AI-native interfaces: Permeable is reimagining attention management through wearables that shield us from distractions based on our context. These design choices show how data can power adaptive interfaces that support without submitting and guide without controlling.
Regulated industries: Healthcare and finance create natural openings that could be potentially too risky for centralized model providers to enter. Health startups like Doctronic are unifying personal health histories across record systems, and we imagine similar plays for financial planning.
This rebel alliance is creating a powerful kind of connective tissue that's structurally incapable of evil because power is distributed, not concentrated in any single entity. Whether these solutions ultimately converge at some point or thrive as an interconnected network matters less than the fundamental shift they represent.
We're looking for more rebels to join this alliance. If you believe in a future where users have access to their digital memory, whether you're building foundational infrastructure or whimsical consumer experiences, we would like to talk.
1 comment
When you front run @usv by a few years. cc @fredwilson.eth https://blog.usv.com/the-rebel-alliance