From a stack of sources
to a single graph.

Start by putting a source into Noosphere. Drop in a PDF of a book or a court filing, paste the URL of a news article, upload a scanned image, or hand it an MP3 of an interview or a podcast episode. Audio is transcribed first, then read like any other text.

Each upload is stored against your account with its own record — who submitted it, when, and where it came from — so every fact the service later asserts can be traced back to the source that produced it.

Noosphere reads what you uploaded and pulls out the load-bearing content: the entities — people, organizations, places, events — and the relationships between them, each with a plain-language description of what connects the two.

Those extractions land in one shared graph rather than a private pile per document. When a second source mentions someone the graph already knows, the service resolves the two mentions to the same entity and attaches the new evidence to it. A name that appeared once in a filing and once in a podcast becomes a single node carrying both citations.

Every node and edge keeps its provenance. Open any relationship and you can see the documents that support it and the sentence each one contributed.

A sketch is a working canvas onto which you pull entities out of the graph and see how they sit together. Search for a person, add them, then expand outward through their connections — the edges that appear are assembled from every document that contributed to them, not just the one you started from.

Sketches are where a graph stops being a database and becomes an argument: you can run community detection to find clusters, follow the shortest path between two entities that never appear in the same document, and keep the canvas as a named, revisitable view. Sketches can stay private, be shared with a group, or be published to a public link.

Most sources turn up while you are already reading. The Noosphere browser extension puts a single button in Chrome, Safari, and Firefox that sends the page you are on straight into your graph — no copying URLs, no switching tabs.

The extension is a convenience accessory. It recognizes you by the same session you use on the website, and the extraction work still happens in the cloud service. Extension downloads and setup.

Noosphere exposes a read-only Model Context Protocol server, so an AI assistant you already use — Claude Code, Claude Desktop, Cursor — can query your graph directly and answer questions in ordinary language: who connects these two people, what do these organizations have in common, which documents support this claim.

You authenticate with the same API token the extension uses. The server is strictly read-only; there are no tools that can write to or alter the graph. MCP server setup.