note.md + ChatGPT (work)

Why this pairing works

ChatGPT's work mode connects to folders and files and shines at the generative end of the pipeline: turning a rough outline into a draft, summarizing a stack of documents, and — increasingly — batch-generating images and diagrams. Left to itself, that output lives in a chat thread you'll lose. Aim it at a note.md vault instead and every deliverable lands as plain markdown (with images beside it), where your reading-annotation loop can catch it.

This is where the "one vault, many agents" idea earns its keep: ChatGPT is rarely your only agent. It's the fast generalist you reach for to produce — and the review, the long-running automation, and the final judgment can each go to whoever's best at that. Same files, different workers.

Setup

1. Keep your vault in a folder ChatGPT can reach — an OpenAI-connected folder, or a cloud/git-synced directory it can read and write.
2. Add an AGENTS.md at the vault root (conventions summary: llms-full.txt) and paste the same house rules into your ChatGPT project instructions — it won't auto-read the file the way a CLI agent does, so tell it.
3. Ask it to save deliverables into the vault as dated markdown, e.g. drafts/2026-07-23-launch-post.md, and to drop generated images into {docname}_files/ with relative links.
4. Open the result in note.md; read, highlight, question — your marks land in sidecar .note.md files, the source stays clean and regenerable.

The loop in practice

You ask ChatGPT to draft a launch post and generate three hero images; it writes drafts/launch-post.md and fills a _files/ folder. You read it in note.md, cut two images, highlight a paragraph that overclaims, and leave a note. Next you hand launch-post.note.md to a more careful reviewer agent — "address the margins." ChatGPT generated fast; the vault kept it; you judged it. That's the division of labor.

FAQ

Does ChatGPT read AGENTS.md automatically?

Not the way a CLI agent (Codex, Claude Code) does. Paste your vault conventions into the ChatGPT project or custom instructions, and point it at AGENTS.md. Sidecars are human-owned except for Smart Lookup's fenced answer writeback; new work should follow your naming rules.

Can ChatGPT-generated images live in my vault?

Yes. Save them beside the document in a {docname}_files/ folder with relative links — the same convention note.md uses for pasted screenshots. They render in the reading view and travel with the vault in git.

Do I have to pick one agent?

No — that's the whole point. Use ChatGPT for fast generation, another agent for careful review, a local agent for private work. They collaborate through the files; you orchestrate. See the orchestration guide.