Import ChatGPT Conversations into Cakumi: A Guide
Import ChatGPT conversations into Cakumi with a downloadable sample, verified parsing results, missing-image checks, and clear memory and cost limits.
Importing ChatGPT conversations into Cakumi involves three separate tasks: obtaining an export, selecting and importing conversations, and checking the result. The transferable material is the readable record. You are not copying a ChatGPT account, all of its platform memory, or the underlying model.
For a long-running character, prepare its instructions and a short description of what you want to continue. Conversation history, character settings, and the feel of future replies need separate checks.
Decide what you want to bring
List a few important chats before importing everything. Record their titles, approximate date ranges, attachments, and the character they belong to.
Keep an untouched copy of the original export. Use a duplicate for inspection or troubleshooting. Renaming a screenshot to .json, or putting copied text into a ZIP, does not turn it into a supported conversation export.
Save your own character description, boundaries, important setting details, and unfinished topics separately. A source transcript does not configure every preference in the destination. Subscriptions and platform-specific features do not transfer with it either.
Request the export through ChatGPT
For an eligible account, sign in and open Settings → Data controls → Export data, then confirm. Follow the notification to download the ZIP and inspect its contents. Consult OpenAI's export instructions for account and managed-workspace availability.
Complete this step with the source platform. Do not give passwords, verification codes, or session tokens to a migration service.
Cakumi's current importer accepts supported ChatGPT conversation JSON and ZIP packages containing conversations and resources. Its instructions also offer conversations.json when that file is present in your export. Keep the complete package when attachments matter.
Export structures can change. Whether a particular file is usable must be established by its preview, not by its extension alone.
Preview and select conversations in Cakumi
Follow the current import flow:
- Open the memory import area and choose Import chat history.
- Select ChatGPT as the source and read the current file-size and processing notices.
- Choose the export file and wait for the conversation preview.
- Select the destination Being, or choose Create Being and complete its details.
- Select the conversations you want and confirm the import.
Previewing does not create a Being or save those chats into it yet, but the file is uploaded for processing. “Not saved into a character before confirmation” should not be read as “processed entirely on your device.”
An unfamiliar structure may use model-based field identification, with usage charges described in the interface. If asked which speaker is you, compare the samples with the original messages. Do not guess from a name alone.
If the file is identified as a character document instead of conversation history, check that result carefully. Importing a persona is not restoration of the full transcript.
An existing Being keeps its current personality and model settings. For a new Being, review the name, description, and instructions yourself rather than expecting the transcript to configure everything.
Download a small sample and inspect what survives
Start with a non-private example before handling a personal archive. Our hand-written sample uses ChatGPT's conversation structure: three messages change a bookshelf-sorting plan from Saturday to Sunday, with a Cakumi logo image and plan.txt. It is synthetic material, not a customer's conversation or an export downloaded from a live ChatGPT account.
- Complete sample ZIP with both attachments
- JSON comparison file without attachment bytes
- Recorded parsing results
On October 21, 2026, we submitted these files to Cakumi's unchanged parsing service running locally in isolation. This checked parsing and recovered attachment bytes. It did not test a signed-in production App session or subsequent model recall.
| Input and mode | Observed result | What it tells you |
|---|---|---|
| Complete ZIP | Recognized as ChatGPT; 1 conversation, 3 messages, 2 attachments; recovered image and text bytes matched the originals | This sample's resources survive when included in the archive |
| JSON alone, missing files not accepted | Missing-attachment error; no successful parsing result | An image reference does not contain the image |
| JSON alone, recoverable-content preview allowed | 3 messages, no recovered attachments, and a missing-file list | A visible placeholder does not mean the picture can be opened |
The sample references its image in two places, and the current missing-file list repeats that image. Check names individually: the displayed number of missing entries is not necessarily the number of distinct files. Keep the original structure rather than deleting fields to make counts match.
If you try the sample in the App, create a separate demonstration Being and check the three messages in preview. Avoid adding demonstration records to a character you already use. This sample helps explain file behavior; your own export still needs its own inspection.
If pictures disappear after import
- Inspect a copy of the original ZIP. Does it contain actual image files that open? A
file-service://…reference, web address, or filename alone cannot restore an absent image. - Check what you selected. The complete ZIP may contain resources stored outside the JSON. Selecting only the JSON leaves those bytes behind.
- Read missing-file warnings. Choose a complete package first. Continuing with recoverable content accepts the gap; it does not regenerate the missing file.
- If imported text appears but an attachment will not open, record the chat, filename, and error, then check sync and connectivity. Remove private information before sharing diagnostics; an entire personal archive is not the first thing support needs.
Check the import before continuing
Use this record:
Source conversation:
Destination Being:
Number of conversations selected:
Titles match the intended conversations:
Beginning, middle, and recent messages are readable:
Speakers are assigned correctly:
Attachments open:
Known missing content or errors:
Character settings or recent context still to add:
Check the record first. Then choose an event actually present in the transcript and explain what you want to discuss next. Asking whether the AI is “the same one” does not establish whether your files were imported correctly.
Saving the conversation and organizing background memories do not finish at the same instant. A successful import does not mean that every detail has become available for recall. Later responses also depend on the Being, chat memory settings, relevance, and model behavior.
For a long history, add a concise new-chat handoff. That gives you a manageable way to check continuity without demanding that the model recite everything at once.
Separate import, memory processing, and chat costs
A saved transcript does not mean that every detail is available for recall. Background memory processing after an active import and continuing the conversation are separate stages. In the current implementation, the first reply in some imported chats without a summary also prepares a history summary; that processing is billed with the turn's actual model usage. Model-assisted identification of unfamiliar files may also incur the usage described on the import screen.
Read the preview's cost notice and select the conversations you need. Before continuing a long imported history, check the selected model and available balance. A successful import does not make future processing of that history free. See the AI chat cost and long-context checklist.
Handle common problems
Missing attachments: check whether you uploaded the complete archive. A remote reference without the actual resource cannot reconstruct an image or document. Choose a more complete file, or explicitly accept importing only recoverable content and record what remains missing.
A parsing or size error: retain the original and check its file type, archive integrity, and the current upload limit. Avoid asking an AI to rewrite the entire JSON just to make it import; you would then need to establish whether it changed the original messages.
Duplicate conversations: imports append chats instead of automatically merging them. After a connection problem, inspect the destination before trying again.
The messages are present but the tone differs: compare model choice and character instructions. Platforms supply different context and settings, so identical behavior is not guaranteed. Use the character consistency guide to make focused adjustments.
Keep the original after migration
There is no need to delete your source record immediately. Retain the original archive, your verification notes, and a separate character document. As new conversations accumulate, update your backups at a pace that suits you.
When evaluating a destination, look beyond a promise of one-click migration. Ask what it imports, what it exports, how failures appear, and how processing costs are explained. The AI companion selection checklist and conversation backup guide help you assess those practical questions.