Anthropic has rolled out one of the most user-centric features in the frontier-model space so far in 2026: the ability to **import memory** from other AI tools directly into Claude.
The new functionality, announced March 3, 2026 and already live on all paid plans (Pro, Team, Enterprise), allows users to transfer long-term context, writing style, project knowledge, personal preferences, role instructions, recurring facts, tone guidelines — essentially their entire “second brain” — from ChatGPT, Gemini, Perplexity, Grok, Llama-based assistants, or any other LLM-powered product.
How Memory Import Works
The process is deliberately simple and privacy-preserving:
1. In your current AI tool of choice, paste the following special prompt (or a localized version provided by Anthropic):
You are now exporting my full memory profile for import into Claude.
Output only a single, well-structured JSON object containing:
- full_name (optional);
- preferred_language;
- writing_style (tone, formality, length, emoji usage, etc.);
- recurring_facts (key personal / professional information I often remind you about);
- project_context (active long-running projects, goals, constraints);
- custom_instructions (any persistent system prompts or rules I gave you);
- important_preferences (formatting, citation style, verbosity, avoidance topics, etc.);
- Do NOT add explanations or markdown. Return pure JSON only.
2. Copy the resulting JSON block.
3. Go to Claude → Settings → Memory → “Import memory from another AI” → paste the JSON → confirm.
Once imported, Claude immediately “understands” who you are, how you like to communicate, which topics you work on, which things annoy you, how detailed or concise you prefer answers to be, and much more — without you needing to re-explain any of it.
Why This Matters
Until now, context was the biggest lock-in factor in the LLM ecosystem. Switching from one model to another usually meant starting from zero: re-teaching writing style, re-explaining your stack, re-listing project constraints, re-setting tone rules.
That friction kept many users stuck even when they preferred another model’s reasoning or speed.
Memory import breaks that lock-in.
Your personal context becomes portable data — a lightweight, user-owned layer that sits on top of whichever frontier model you choose to use next month or next year.
Anthropic is explicitly framing this as the beginning of a larger architectural shift:
“The model is becoming commoditized.
What remains scarce and valuable is the user’s long-term context.
That context should belong to the user — not to any single company.”
Competitive Landscape & Reactions
- OpenAI has offered “Custom instructions” and “Memory” for over a year, but both are locked inside ChatGPT. No export.
- Google Gemini has Workspace / Notebook memory, but again — no clean export format.
- xAI Grok has very little persistent memory at all.
- Perplexity and most RAG-based tools treat context as ephemeral.
Anthropic is the first major player to make memory exportable and importable in a standardized, documented way.
Early user feedback on X and Reddit (as of March 3 evening):
- “Switched from GPT-4o to Claude 3.7 in 47 seconds. It already knows my entire Next.js + Tailwind + shadcn/ui stack.”
- “This is the first time I actually believe I can change models without losing months of accumulated context.”
Also read:
- Former GitHub CEO Nat Friedman Unveils Entire: Revolutionizing AI Development with Git Integration
- Andon Labs' AI Office Manager Bengt Hires a Human: A Step Toward AI-Human Collaboration in the Physical World
- Big Tech's $650 Billion AI Arms Race: Four Companies Outspend Entire Nations on Infrastructure
Availability & Limitations
- Available today on all paid Claude plans (Pro $20/mo, Team $30/user/mo, Enterprise custom).
- Free users do not have access (yet — Anthropic says “working on lighter version”).
- Maximum imported memory size is currently 128 kB of JSON (roughly 60–80 k tokens of context once unpacked).
- Anthropic promises no training on imported memory and end-to-end encryption during transfer.
Official documentation and ready-made export prompts for every major model are available at:
→ https://claude.com/import-memory
The quiet launch of memory import may turn out to be one of the most strategically important moves of early 2026. In the race to commoditize frontier intelligence, the company that first makes user context truly portable — and therefore truly owned by the user — could reshape user loyalty for years to come.
