Skip to content
agentics.you
Start free

Docs

Import what your assistants already know

Bring what your other assistants already know into your workspace memory: a Claude memory export, a ChatGPT data export, or your notes. You review every fact before anything is saved, and each saved fact is marked as coming from outside your agents.

What you can import

  • Claude: its memory export, memories.json, or the Claude export .zip that holds it. Each sentence of the memory summary, and of each project's memory, is a fact to review.
  • ChatGPT: its data export, the .zip from ChatGPT's settings, or the conversations.json inside it. The facts are what ChatGPT itself remembered: its memory entries, the memory list a conversation started with, and your custom instructions. Your chats themselves are not read for facts.
  • Notes: Markdown and text files (.md, .txt), a folder of them, or a .zip, including folders exported from Notion and Obsidian vaults. Each list item and paragraph is a fact, tagged with its note's title and headings.

Anything else is refused with an error that says what import reads. It never guesses at a format. One import takes up to 50 MB and up to 10,000 facts. A bigger one is refused, never cut short, so import it in parts.

Import in the console

  1. Open Import in the console and choose Assistant memory. The facts go to the namespace chosen on that page: the default one unless you pick another.
  2. Choose what you are importing, or leave it on Detect, and add the files.
  3. Choose Read and review. The export is read on your device by the same parser the server uses, so a large export never leaves it: only the facts it finds are sent.
  4. Review the facts, and untick any you do not want. A fact that looks sensitive or reads like an instruction is flagged, and left out unless you tick it.
  5. Choose Save. The facts are written in batches, with a progress bar.

An import you do not confirm within 72 hours expires, and its facts are deleted. Cancel import ends it at once; facts already saved stay. Importing needs an owner or admin in the workspace, and imported facts do not count as actions on your plan.

How imported facts are stored

Each fact you save is a memory with a memory.imported receipt. It is stored with trust_level external and tainted true, and recall returns it with those fields and untrusted true, so an agent can tell it from what your own agents wrote. A fact that is already stored is counted, not saved twice.

With the API

The console makes these calls, and so can you, signed in as an owner or admin or with a key that has the admin scope. The REST API reference has every field.

  • POST /v1/ingest with kind import starts an import from the files (up to the 8 MiB request body) or from what your own parser found. It answers 201 with the import waiting for review, and writes nothing yet.
  • GET /v1/ingest/imports/{id} returns the import and its facts, for review.
  • POST /v1/ingest/imports/{id}/confirm takes your choices, include for flagged facts to keep and exclude for facts to leave out, then writes up to 50 facts. While remaining is above 0, call it again with no body.
  • POST /v1/ingest/imports/{id}/cancel ends an open import.

An unknown format answers 415 unknown_format. An upload over 50 MB answers 413 upload_too_large, and one with more than 10,000 facts answers 413 too_many_facts.