AI Soul Archive Explained: What It Is and Why You Might Need One

An AI soul archive is a small portable file — or short set of files — that describes who you are in a format any AI can read. You keep it on your own device. You paste it into ChatGPT, Claude, Gemini, or a local model when you want the AI to actually know you. The AI then responds in your voice, with your context, without you having to explain yourself every time.

That's the entire concept. The rest of this article explains why it's becoming a real category in 2026, what's in a good archive, and why no AI company will build one for you.

Why "AI Memory" From OpenAI Isn't an Archive

ChatGPT, Claude, and Gemini all advertise some form of memory. None of these is actually an archive. They are short, server-side note stores attached to your account, with three structural problems.

A soul archive solves all three. The file is yours, the original is preserved, and the only cap is the AI's context window — which keeps growing every year.

What's Inside a Good Archive

There's no industry standard yet, but the practical structure that's emerged from people who actually use this pattern uses six to ten markdown files. Each file has a specific role.

• SOUL.md — Narrative identity. Voice, values, what you keep returning to, how you write. 1,000-5,000 words.
• MEMORY.md — Structured facts. Background, projects, preferences, current goals. 500-2,000 words.
• MY_CANON.md — Your body of work. Books, essays, products, the artifacts you've made that matter.
• MY_PORTALS.md — Where you publish or transact. Domains, profiles, brand presences, audience funnels.
• MY_ARTIFACTS.md — Concrete things you've shipped: products, code, designs, recipes, anything tangible.
• MY_OPERATIONS.md — How you work. Daily rhythms, weekly cadences, decision rules.
• MY_REVENUE_MAP.md — How money moves through your life or business.
• MY_RED_LINES.md — What AIs are not allowed to do for you. Where your taste is non-negotiable.
• MY_ARCHIVE.md — Closed projects worth remembering but not currently active.
• MY_FIRST_7_DISPATCHES.md — The next seven things you want to ship, in order.

Not everyone needs all ten. A writer might only use SOUL.md and MY_CANON.md. A solo founder might use all ten. A person in a research role might use SOUL.md, MEMORY.md, and MY_OPERATIONS.md. The system flexes.

What an Archive Actually Changes

Once the files exist on your device, the daily friction with AI drops sharply.

• New ChatGPT session — paste SOUL.md. ChatGPT responds in your voice immediately. No more "I'm a writer, I prefer concise replies, I'm working on…" for the hundredth time.
• Trying Claude for the first time — paste the same file. Claude understands you within one message. The investment in ChatGPT transfers.
• Asking AI to draft a pitch / email / blog post — paste SOUL.md + MY_CANON.md. The output sounds like you wrote it, because the AI has your actual voice and the context for what you've made before.
• Hiring a new AI tool — Perplexity, Copilot, Cursor, Notion AI, a future model. Paste the file. Each new tool starts with full context instead of a cold start.
• Account closed, pricing changed, model deprecated — you don't lose anything. The file is the constant.

The archive is not what an AI remembers about you. It is what you keep, in case the AI forgets.

How to Build One (Without Writing 10,000 Words)

You can write the files manually. Open a markdown editor, create the file, type the sections. Plan for 4-8 hours of focused work for a complete archive. People who keep journals find this easier; the journal already contains most of the SOUL.md material.

For people who'd rather not spend a day on it, there are now tools that read your existing writing and generate the structured files automatically. The pattern is: paste 5,000-50,000 words of your own writing into one input box (blog archive, essay collection, long emails, public bios, social posts), and the tool returns all ten files in one pass.

Soul Alchemy is one such tool, built specifically for this output format. The cost is one-time ($99), not subscription, and the files are yours to download and edit forever. The AI doing the generation never retains your input. It's the cleanest end-to-end path from "I have writing scattered everywhere" to "I have a portable identity any AI can read."

Privacy: What to Put In and What to Leave Out

The file is as safe as what you put in it. Most users keep two versions:

• Private archive — full version, with addresses, family details, medical history, financial information, anything that makes the file complete but sensitive. Stored locally, encrypted at rest, never pasted into a public AI.
• Public archive — redacted version. Removes anything you wouldn't post on a personal homepage. This is the file you paste into ChatGPT, Claude, Gemini, or any company-owned model.

Treat the file like a passport: useful, valuable, not for posting in public places. Opt out of training data collection at the account level for any AI company you paste into (OpenAI, Anthropic, and Google all have this setting). Then paste with confidence.

Why This Matters in 2026 Specifically

The category exists now because three things came together in late 2025 and early 2026.

First, context windows got long enough. Claude 4.6 and Sonnet 4.7 ship with 1M token windows. GPT-5 has 400K. You can paste a 50,000-word archive at the start of a conversation and the AI holds it for the entire session. Two years ago this was impossible; the file would have been truncated before the AI saw the bottom.

Second, people started caring about lock-in. The "switch tax" from leaving one AI is no longer abstract. People who used ChatGPT for 18 months and tried to move to Claude noticed exactly how much they were leaving behind. The market started looking for portable formats.

Third, the markdown convention solidified. SOUL.md and MEMORY.md are now common enough patterns that any AI recognizes them when pasted. The format is not patented or owned — it's just a small, robust standard that emerged from practice. That permanence is part of why it works.

Frequently Asked Questions

What is an AI soul archive?

An AI soul archive is a portable file (or small set of files) that describes who you are in a format any AI can read. The most common format is markdown: SOUL.md for narrative identity, MEMORY.md for factual recall, and structured files like MY_CANON.md, MY_PORTALS.md, MY_RED_LINES.md for specific aspects of your work and values. You own the file. You paste it into any AI conversation. The AI knows you immediately.

Why can't ChatGPT or Claude memory replace a soul archive?

ChatGPT memory belongs to OpenAI. Claude memory belongs to Anthropic. Both store short summaries on their servers, both have capacity limits, and neither follows you when you switch tools. A soul archive is a file you own on your own device. It works across ChatGPT, Claude, Gemini, Perplexity, local models, and any future AI — because the file is yours, not the company's.

What does a soul archive contain?

A typical archive has six sections: Identity (who you are and how to address you), Voice (how you write and what you sound like), Canon (the body of work you've made — books, essays, products), Portals (the places where you publish or transact), Red Lines (what AIs are not allowed to do for you), and Archive (closed projects worth remembering). The file is usually 1,000-5,000 words and takes 4-8 hours to write by hand.

How do I build a soul archive without writing it from scratch?

Tools like Soul Alchemy read your existing writing (essays, journals, blog posts, emails, public bios) and generate the structured files automatically. Paste 5,000-50,000 words of your own writing in one input box, and the tool returns SOUL.md, MEMORY.md, MY_CANON.md, MY_PORTALS.md, MY_ARTIFACTS.md, MY_OPERATIONS.md, MY_REVENUE_MAP.md, MY_RED_LINES.md, MY_ARCHIVE.md, and MY_FIRST_7_DISPATCHES.md in one pass. The output is yours to download and edit forever.

Is an AI soul archive safe to share with an AI?

The file is as safe as what you put in it. Most users keep the personal version private and create a redacted public version for AI sessions (no addresses, no full names of family, no medical history, no financial details). Treat the file like a passport: useful, valuable, and not for posting in public places. The OpenAI / Anthropic / Google terms of service are clear that they may use chat content for training unless you opt out at the account level — opt out, then paste.