Agents today are largely blind to the user. Even when they add memory, they usually start from zero instead of using the rich context already locked up in consumer platforms. The result is agents with big gaps in their understanding of the user, generic experiences, lots of repetitive input, and still poor engagement.
We’re trying to solve this by focusing on *portable, user-controlled personal context*. **
With Fabric, an agent doesn’t just recall what happened inside that one app. It can start from a richer baseline that evolves as the user interacts across their existing apps, using memories created from tens of thousands of real interactions, such as:
- Travel posts and restaurant stories from Instagram - Videos watched on YouTube and TikTok - Years of Google search, shopping, and navigation behavior
What we’ve built so far
- Official connectors for Instagram (global), Google (EEA, UK) and YouTube (EEA, UK) for beta users. First company globally to build DMA style integrations with all Big Tech platforms. - An MCP server so users can bring their context into MCP clients - Under the hood, Fabric: - Ingests raw interactions that look very different (e.g. Google searches vs Instagram stories) - Normalises them into an ActivityStreams-style schema - Builds a personal knowledge graph on top - Exposes higher-level “memories” that agents can retrieve via our MCP server - A user portal where users can connect data sources and manage their context
We have a few strong opinions:
- The profile belongs to the person, not to any one app. Fabric is designed as a neutral “personal context vault” that multiple AI apps can plug into (with user approval), rather than a proprietary profile owned by a single product. - User control is first-class. Users can: - Choose which data sources to connect - See which apps/agents have access - Revoke access and delete their context from Fabric - We’re explicitly building on top of GDPR data portability and EU/UK “smart data” initiatives, rather than shadow integrations.
We would love to hear what you think.
And if this is interesting, you can sign up to our waitlist or join our beta at onfabric.io.
No comments yet.
