01The AI memory tax
The promise of AI is that it will understand you, learn from you, and help you more effectively over time. That promise is not being kept.
Every new conversation starts from zero. ChatGPT does not know what you told Claude. Claude does not know what you told ChatGPT last Tuesday. Gemini does not know what your teammate spent three hours researching yesterday. Every AI app lives in its own silo, with its own isolated memory, with no awareness of the broader ecosystem you live in.
The result is a tax on the user. You re-explain your context. You re-introduce yourself. You repeat your preferences, your goals, your constraints — over and over, to app after app, session after session. You are the only glue holding your AI stack together, and you are doing it manually.
The tax compounds. A power user switching between five AI tools throughout the day loses context at every transition. A new user signing up for a promising new app immediately hits a wall — the app has no context about them and feels generic until weeks of use accumulate. Most users never make it that far. A small team researching the same topic across multiple sessions has no shared memory infrastructure. Team members duplicate effort without knowing it. Research disappears into individual chat histories that nobody else can access.
02Memory as a feature is a dead end
The deeper problem is structural. AI apps are built as endpoints — interfaces for a conversation. Memory was an afterthought, tacked on as a per-app feature rather than designed as shared infrastructure. There is no standard for how AI apps should store, share, or retrieve knowledge about a user. Each app reinvents a limited version of memory independently.
Per-app memory will always fall short for the same reason every walled garden falls short. It optimizes for the app, not the user. It locks context inside the tool that captured it. It cannot help when you open a different app tomorrow. The more apps you use, the more fragmented your context becomes — even if every individual app's memory works perfectly.
This is not a missing feature. It is a missing layer.
03The case for a layer
Memory belongs above the apps, owned by the user.
Think of Index as the nervous system between your AI apps. As you use them naturally, they continuously write structured knowledge into your Index — organized into visual, topic-based spaces called Boards. When you start a new session with any connected AI, that AI pulls from your Index first, arriving with full context already loaded. The more you use your AI tools, the smarter and more contextually aware every one of them becomes.
The memory is not owned by any AI app. It lives in Index, belongs to the user, and is accessible by any connected tool. What you tell ChatGPT is available to Claude. What your colleague researches with Gemini is available to your Claude. The layer is AI-agnostic by design — and the user is at the center of it.
04What the layer looks like
Index is a small number of pieces, each one doing one thing well. See features for the long version.
- Boards. The fundamental unit of memory. Each Board is a named, organized space containing everything Index knows about a specific topic, project, interest, or domain. Personal or team-shared. Living, not static.
- The Orchestrator. The central nervous system. Every piece of information that enters Index passes through it: classify, route, dedupe, merge, decide whether to trigger the Proactive Engine. Fast deterministic logic for the easy cases, a lightweight LLM for the ambiguous ones. Every decision logged and inspectable.
- The Memory Engine. Semantically indexed storage. Deduplicates and compresses as Boards grow. Returns the most relevant entries to a connected AI on demand — ranked by semantic similarity, recency, and frequency of reference.
- The Proactive Engine. Continuously analyzes Boards for patterns, trends, connections, and actionable signals. Surfaces what it finds as ambient cards on the dashboard. Never pushes. Never interrupts.
Connectors hold the layer to the outside world. Each Connector has two capabilities: write (the AI app pushes knowledge into Boards) and read (the AI app retrieves context before responding). Permission-scoped, revocable, transparent. Index provides official Connectors for the major AI apps. Anyone can build their own.
05The compounding effect
Once memory is a layer, every new connector inherits all prior memory.
Connect a new AI app to Index and it immediately has access to your Boards — months of accumulated context about your goals, decisions, prior research, and open questions. The new app is useful from the very first message. No cold start. No warm-up period. The onboarding wall that kills most new AI apps simply does not exist for tools connected to Index.
Inside a team, the same effect plays out at a different scale. Every team member's AI activity enriches the shared Boards. A team of five using AI tools intensively produces a Team Board that is five times richer than any individual's. Research becomes additive rather than redundant. No valuable context gets siloed in an individual's chat history.
Knowledge compounds across tools, across teammates, across time — automatically, without manual effort, without repetition. The whole AI stack learns together.
06What we won't do
Some things are out of scope on purpose.
Index is not a chatbot. It does not generate responses, complete tasks, or replace your AI apps. It makes all of your AI apps better by solving the one problem they share: they forget. Product decisions that would turn Index into a generic AI interface are out of scope.
Index does not surveil users. The optional Passive Learning Module is permission-gated and inspectable. Raw screenshots are never stored or transmitted. Only extracted signals — topic keywords, domain categories, frequency data — are processed and written into Boards. The user can view the full log, pause the module, exclude apps or domains, and permanently delete all passively collected data.
Index does not train on user data. All data stored in Index belongs to the user, not to Index. Users can export, delete, or revoke access at any time. We do not share data with third parties beyond the named sub-processors required to deliver the service.
Index does not intrude. The Proactive Engine surfaces what it finds as ambient cards on the dashboard. No push notifications by default. The user engages on their own terms.
07The bet
Index is positioned to become the standard memory protocol for AI apps — the way that OAuth became the standard for authentication.
The pattern is familiar. A capability that was previously reinvented per-app becomes a layer, and the layer becomes infrastructure. Once the layer exists and works, every new app ships with it built in — because not shipping with it is worse for both the app and the user. Users stop tolerating tools that ignore the standard. Developers stop pretending they can do it better in isolation.
That is the bet. Memory as infrastructure. Index as the layer. The user at the center, in control of their own memory, with every tool they choose to trust building toward something larger than any one of them.
Read the roadmap for how we get there.