Pitch — for Builders / Collaborators

The version tuned for engineers, designers, researchers, and co-conspirators evaluating whether to work on IdeaFlow. Leads with substrate, not shipping.

The bet

First build a medium for thought, then there are things you can build in world with it. — Joe Henke, IdeaFlow, early 2024

IdeaFlow is infrastructure-first. The substrate is the point; the applications compose.

See ../framings/medium-for-thought.

What the substrate is

A graph data model expressive enough to hold text, relations, spatial regions, category-theoretic structures, calendar events, people, conversations, files. The in-house framing: the God data structure / Gopher backbone / GraphMirror (Jacob, 2025: "Everyone uses ProseMirror. We could offer GraphMirror.").
An editor that feels like workflowy but is a graph (Taylor Mitchell's 2024 framing: "graph database explorers expand all related nodes, not just children").
LLMs doing the markup burden so users don't have to.
Federation-ready from day one — each user's graph is a subgraph of a larger commons.

Open-source movement

Jacob, 2024-10:

"Starting the open source movement for knowledge graphs, e.g. Open Supply Chains."

This isn't a walled-garden KM product. The long-arc goal is a commons of structured knowledge with federation and credible-neutrality properties. See ../framings/open-source-knowledge-graphs.

The interesting problems

Graph editor UX. No one has shipped a truly graph-native editor that doesn't bounce off mainstream users.
LLM + structure feedback loop. Markup without burden means LLM-generated structure with human-in-the-loop review. Review UX is wide open.
Federation without extraction. How do personal graphs federate into team / public graphs without siphoning value away from their authors?
Hypergraph-store-for-LLMs. Jacob, 2025: "It helps language models … do this equipping them with a hypergraph store." Interesting adjacent research thread.
Provenance at every node and edge. The meeting-notes demo (Oct 2025) claims "the first well-organized meeting notes ever taken in human history" because of provenance. Making that default, not artisanal, is a substrate problem.
Chunking a God data structure. How do you make a data model capable of everything usable for one thing?

Why now

LLMs close the Semantic-Web markup gap.
The tools-for-thought movement has run into its ceiling without structure.
Hypergraph / property-graph databases are fast enough at scale.
The AI-agent wave needs a persistent substrate that survives any single model.

Lineage

Doug Engelbart → Jack Park → Jacob.
Tim Berners-Lee (Jacob's Oxford/MIT advisor).
Vannevar Bush's memex as the patron saint.

What we value in collaborators

(Inferred from the Slack corpus, not a formal HR statement:)

Care about the two-stage structure — medium before applications.
Comfort with both graph abstraction and UX intuition. Taste matters.
Willingness to say "this framing is wrong" when it is. (Taylor and Cody did this repeatedly and moved the product.)
Appreciation for speculation as a work product (see ../open-questions).