Why a library

Updated

You are repeating yourself to AI, and what it learns, it won’t share.

Every task needs the same context: the job, the rules, the data, the examples that worked before. It gets rebuilt by hand every session. Long sessions grow too large and start to rot. The work ends up in chat transcripts nobody rereads, and what comes out is a deliverable, not something reusable or handable to a colleague. Assistants offer project folders and memory, then rent them back. A Claude project goes stale the moment a file changes, and what Claude Chat knows, Claude Code can’t read.

FactorIn’s answer is that LLMs work better in files than in chat.

What a library is

A library is a folder of plain-text files: notes, prompts and skills, plus the attachments they refer to. Some you write, some you borrow, some you generate from the body of work you already have. It reaches three ways.

  • Every AI app. Claude, ChatGPT, Cursor, Claude Code, Codex and anything else that speaks MCP reads and writes the same files through one connector URL.
  • Every device. The library is the same wherever you open it: in the web app, or in any AI app connected to it.
  • Everyone you work with. One shared library for the team, a personal one per member, and a record of what gets used.

Why files

  • A task is a file. How to do it is links to other files: the style guide, the prior art, an agent skill. Developers call this spec-driven development. FactorIn carries the practice beyond code.
  • Every task gets a fresh context. The files you chose and nothing else. No rot, no leftover state from last week’s session.
  • Nothing sticks without your say-so. Everything written to the library is a version you keep or roll back. Every save records what produced it: which app, which run, which copy it came from.

What you get that chat memory does not

Portable. The files are markdown and ordinary attachments, in folders with the names you gave them. The whole library exports as a ZIP in one click, and what comes out is a folder you can open anywhere. A library whose plan lapses goes read-only and still exports whole. AI keeps changing; your library doesn’t.

Not proprietary. The connector is the open MCP standard that Claude and ChatGPT already use. A skill is the same SKILL.md folder Claude and Codex read natively. A prompt is a markdown file with optional JSON Schemas for its inputs and output. Nothing here needs FactorIn to read it back.

Versioned. Every save of every file is a version with a diff and a rollback. Archiving hides a row; it never destroys one. A wrong edit by an AI app is one click from undone.

An audit trail. A version knows what wrote it. On the Cloud plans, every asset a run produces records the prompts, the models and the cost that produced it, step by step.

Shareable, and private by default. Nothing in a library is public. A file is reachable through a membership or a guest grant, or it is not reachable at all. A guest sees one folder and everything under it, read-only, and gets a dropbox of their own to write in. A guest can copy what they read into their dropbox, and the copy remembers where it came from.

Borrowable. A folder can mirror a GitHub repository, pinned to a branch, tag or commit, optionally pulled nightly. Skills and reference material enter your library without being retyped, and the mirror root carries the repository’s license.

What it is not

It is not a chat app, and it does not replace the Claude or ChatGPT seat you already pay for. The library appears inside that seat. It is not a sync-and-editor product either: Obsidian users looking for a vault host will find it a poor fit, and that is deliberate.

Who it is for

Small teams that already work with Claude or ChatGPT together, bought by the one member who sets up the tools. A marketing or content duo with a brand voice to keep. A founder-led operations team with playbooks, digests and investor updates. A small product pod building with Claude Code or Cursor. If you have a Monday plan, a Thursday digest, or a build session that starts with re-teaching the agent its rules, you have a library already. It is just scattered across chat windows.

Next: connect Claude, connect ChatGPT, or connect another app.

Start my library