Sofia, from Intuition Box, is an AI-powered knowledge management system that captures a user's web browsing activity and converts those signals into verifiable, on-chain knowledge. The app records web sessions through a Chrome extension, applies AI workflows for intent classification, and exposes personalized context to models via a Model Context Protocol server. A web dashboard surfaces attestations and reputation data. Developers and Web3 users gain a local-first path to make browsing-derived knowledge usable for agent-driven workflows.

What tasks can you actually use it for?

Sofia turns passive browsing traces into structured artifacts that agents can query, producing what the developer calls certified knowledge on a decentralized protocol. That output targets workflows where model responses must reference a user-specific knowledge graph, for example contextual query answering, personalized recommendations driven by browsing history, and on-chain reputation tracking. The system emphasises converting ephemeral web signals into attestations that downstream agents can reference.

How verifiable and dependable are the generated attestations?

The app produces attestations anchored to the Intuition protocol, so verifiability rests on decentralized attestations and on-chain reputation records rather than only local metadata. AI-driven workflows classify intent and structure captured items before attestation, which means the factual reliability of a certificate depends on the quality of captured pages and the classifier's labels. For high-stakes decisions, the generated attestations require independent review and cross-checks.

What inputs and integration requirements matter in practice?

Sofia accepts live web browsing as its primary input, and the stack requires Google Chrome for capture, an MCP-compatible client (for example, Claude Desktop) for server-side model access, and a modern browser for the dashboard. The monorepo does not advertise native ingestion of arbitrary file formats, so non-browser sources need pre-processing or separate tooling. Dependence on specific clients limits out-of-the-box compatibility for some workflows.

Does it fit into developer workflows or demand specialist setup?

The project is delivered as a Bun-based full-stack monorepo intended for developer deployment, and it exposes an MCP server to localize model context to a user's knowledge graph. That architecture favours engineers familiar with Web3 identity and attestation concepts, and it integrates with on-chain reputation systems surfaced by the dashboard. Teams should allocate integration effort to connect MCP clients and to map attestations into existing agent policies.

A pragmatic choice for Web3 teams that plan governance and verification

Sofia is a pragmatic option for developers who need verifiable, browsing-derived context available to AI agents, provided teams commit to governance and verification workflows. Before granting agents broad access to attestations, audit the classified records and define permission boundaries for on-chain reputation. Treat generated certificates as machine-assisted evidence that requires human checks and policy controls in production deployments.