Product — AI & MCP

Hand the model your structure

Put AI to work on the particular shape you've given your material. Our MCP server exposes your sources, notes, links, writing, and past searches to the tools you already use — coding agents, RAG pipelines, custom agent loops. The model works from your context, not a flattening of it.

You · May 29, 2026

Curation isn’t storage — choosing and arranging IS the interpretive act. The shelf is a first draft.

Linked sections

You · May 27, 2026

Cf. Dewey: reflection as the deliberate re-ordering of experience — the order is the argument.

in source Attention, Curation, and the Shape of Thought

Memory as Reconstruction

Added by Jake · yesterday

Remembering has long been modelled as playback: a record retrieved intact from storage. The experimental record reads otherwise.

Recall is performance, not playback — each act of recall is a small act of revision — the remembered scene is rebuilt around the concerns of the present.

In this sense what is stored is not the experience itself but the most recent reconstruction of it. The archive is alive, and every visit leaves fingerprints.

For shared work the implication is immediate: a collaborator returning to your notes reconstructs them through their own questions — division of attention becomes division of memory.

None of this argues against keeping records. It argues for keeping them deliberately — knowing the keeping itself will do quiet editorial work.

Linked sections· drawn Apr 14

Memory as Reconstruction

each act of recall is a small act of revision — the remembered scene is rebuilt around the concerns of the present

Attention, Curation, and the Shape of Thought

We tend to imagine recall as retrieval, as if the past sat unchanged in a drawer. But the file you reopen has been quietly edited by everything you filed beside it.

does recalling a memory change the memory itself

Attention, Curation, and the Shape of Thought
1 matching section
- We tend to imagine recall as retrieval, as if the past sat unchanged in a drawer. But the file you reopen has been quietly edited by everything you filed beside it.

~/cognition

~/cognitionmcp · thought-process ✓

A coding agent, recollecting over the library

What you can do

An MCP server

Sources, notes, links, writing, and past searches — exposed as tools any MCP client can call.

Bring your own model

Connect Claude Code, Cursor, a RAG pipeline, or any agent loop — and swap models whenever you like.

Built-in chat

Team plans include chat over your sources, notes, and search — no setup required.

Never trained on

Your material is never used to train models. Third-party AI runs under your own agreement with that provider.

In practice

Recollect, summarise, and synthesise across everything you've kept
Follow the links and citations you drew — the model inherits your structure
Works from your terminal or IDE, over the same library your team shares
The model is yours: its data handling follows your plan with that provider

Questions

Is my material used to train AI models?

No. Your uploaded sources, notes, and writing are yours alone and are never used to train models. Integration with third party AI is up to you and governed by your usage agreement with those models.

Which AI models can I use?

Whichever you bring. Thought Process exposes your library through an MCP server, so you can connect coding agents like Claude Code, Codex, Cursor, a RAG pipeline, or any agent loop — and swap models whenever you like.

Build your library today.