You back it up. I rebuild it. You get it back, better organized than it ever was.
7–10 business days, start to finish, depending on the size of your backup.
The archive and your identity travel on separate tracks. From step 3 your archive moves under a random ID with the files that name you sealed off. Your record keeps only a keyed fingerprint of that ID, and the two are matched with a key that never touches the work machine, at step 7, after testing passes.
The seven steps
Step 1
You export your backup.
Every major AI platform lets you download your history.
I’ll show you where the button is for yours.
The file you get is yours to keep whether or not you work with me.
Step 2
You reserve your place and send the file.
A reservation takes a $497 deposit.
This pays for a Basic Transfer in full and counts toward any bigger job. After that, I email you instructions for sending the backup. The file is what I work from; nothing is touched inside your existing account.
Step 3
Your archive gets a blind ID.
Think of a warehouse. A pallet comes in and gets a tracking tag, not a name.
The moment your file finishes uploading, it gets a long random processing ID (something like K7QX-M2PD-9VHT-4RZA) that becomes its name for the entire job. The parts of the export that say who you are, like the account file with your name and email, are sealed off before any work starts. Your reservation record never stores that ID either, only a keyed fingerprint of it. Someone holding the archive can’t find you, and someone holding the customer list can’t find the archive.
Step 4
I sort it by what it actually is.
A chat history is a pile. A brain is a filing system.
I separate the pile into decisions you’ve made, facts about your life and work, questions still open, and preferences, and I keep track of where each one came from. That’s provenance: you can always ask why does my agent believe this? and get an answer.
Step 5
I rebuild your agent in its new home.
A private environment under your control.
With its personality, instructions, and memory set up the way they were in the cloud. Then upgraded, because the memory is sorted now, not just copied. A Basic Transfer runs on Honcho; on the Memory Upgrade and Multi-Agent tiers you choose the memory system (more below).
Step 6
I test it.
Purely technical: are the memories in the right places, does the agent find what it should, do the instructions load. I’m checking structure, not reading your life. (More on that below.)
Step 7
You meet your agent again.
Only after the tests pass do I use a key, kept off the work machine, to match the processing ID back to your reservation. You get a message that your agent is ready, and then the keys are yours. Thirty days after you have it, the link between that ID and your name is deleted, so the match can’t be made again.
What “blind” means, exactly
It’s a process, not a promise.
Your archive is handled under a random ID the whole way through, with the files that name you sealed off, and it is matched to your name only after the rebuilt agent has passed testing. The match takes a key that lives on the intake side, not on the machine doing the work. Nobody has to trust a vague claim that “nobody looks.” The order of operations, and where the key lives, is the safeguard.
What it doesn’t mean: the data is invisible to the person doing the technical work. To rebuild a memory system, the work has to handle the content, and your conversations may mention your name on their own. That’s why the ID matters. The content is processed without a name attached to it, and nothing on the work side points back to you.
The upgrade, not just the move
Most people think of this as a rescue: get my stuff out of the cloud.
It is. But the cloud never gave you a memory that was organized.
It gave you a long scroll that sometimes remembers a few things.
When I rebuild your agent, the memory is sorted into decisions, facts, open questions, and preferences, with a record of where each came from. That’s something an AI can reason over, rather than just search through. For many people the upgrade is the bigger win. The move is the price of getting it.
Kinds of brains
Not every agent needs the same kind of memory.
A Basic Transfer runs on Honcho. On the Memory Upgrade and Multi-Agent & Teams tiers you choose, and I’ll recommend one based on what you bring me, and tell you why.
Honcho.
Built by Plastic Labs, Honcho focuses on modeling the person: it builds and refines a picture of you over time from your conversations, rather than just storing them. Best for agents that act as an ongoing assistant or collaborator, where understanding how you think matters more than lookup. It’s released under the AGPL-3.0 open-source license. This is the Basic Transfer default.
Open Brain.
A database-backed memory (built on Postgres with vector search, reachable by AI tools through the Model Context Protocol, a standard way for AI apps to connect to outside tools). Best when you have a large archive and want your agent, and your other AI tools, to search across years of material quickly. Created by Nate B. Jones.
OpenViking.
From Volcengine, OpenViking treats memory like a filesystem: folders and files the agent navigates, with three levels of detail (a short summary, an overview, and the full content), so the agent reads only as deep as it needs. Best for very large or highly structured archives, such as a business’s documents and project history. Also open source (AGPLv3).
Also available on the upgrade tiers: Mem0, the most widely used drop-in memory layer; Letta, a full agent runtime where the agent manages its own memory; and Graphiti (the open-source engine behind Zep), which tracks how facts change over time, so your agent knows what used to be true as well as what is true now.
You aren’t locked in to my choice: because memory is stored in open, documented formats, it can be moved again.
(Every option is open source. That’s the point: the brain should be something you can inspect and take with you.)
Common questions
What if I only have a little history?
That’s fine. A small, well-sorted memory still beats a big unsorted one, and it grows from there.
Does my existing account get touched?
No. I work from the file you send. Your cloud account stays as it is until you decide otherwise.
Do I need technical skills?
Step 1 is clicking an export button. Everything after that is my job.
