Browsex

A semantic, local-first web browser that accumulates knowledge over time.

Vision

Traditional browsers treat pages as ephemeral documents. Browsex treats pages as sources of claims about the world.

Each visit:

• Extracts semantic assertions (JSON-LD, RDFa, Microdata)
• Records their provenance
• Links them to previously seen entities
• Adds them permanently to a growing local knowledge graph

Over time, the browser becomes a personal semantic index of the web — a research memory and substrate for agentic reasoning.

Technology Stack

• Elixir / BEAM — concurrency, supervision, long-lived memory
• Elixir Desktop — cross-platform packaging (Windows, macOS, Linux)
• Phoenix LiveView — reactive UI and semantic side panels
• Embedded WebView — real web rendering engine
• RDF.ex — canonical RDF data model
• Jido — agentic workflows and reasoning

Architecture Highlights

RDF as Primary Model

All semantic data is stored as normalized RDF quads (graph, subject, predicate, object). Named graphs provide first-class provenance tracking, enabling selective forgetting, trust weighting, and conflict analysis.

Semantic Enrichment

When pages lack embedded RDF, the browser can:

• Extract visible text and identify candidate entities (NER)
• Resolve entities against knowledge bases (Wikidata, MusicBrainz, DBpedia)
• Fetch and display RDF for matched entities
• Optionally highlight detected entities in page content

CRDT-Based Sync

Knowledge graphs use conflict-free replicated data types:

• OR-Set for shared graphs (add-wins, safe collaboration)
• LWW for private graphs (last-write-wins, simple editing)

Operations stored in a Merkle-DAG for distributed sync.

Status

This project is in the research and design phase. See notes/research/ for specifications:

License

MIT — see LICENSE.md