Where Your Chat History Actually Goes When You Export It: A No-Fluff Look at JSON Files, Embedding Vectors, and What You Can (and Can't) Reimport to Another App

The 30-second answer

When you export your chat history from an AI companion app, you get a zip file containing JSON-formatted message logs, some metadata about your companion's personality settings, and a set of high-dimensional embedding vectors that represent the semantic meaning of your conversations. You can open the JSON in any text editor, but those vectors are useless outside the original app's vector database, and no mainstream AI companion app supports importing another app's export format. Your data is yours, but it's not portable.

What's actually in that zip file

You hit export. The app sends you a download link. You unzip it and see a folder with a few JSON files, maybe some audio clips if you used voice mode, and a metadata file that looks like machine-generated nonsense. It's not nonsense. It's a structured snapshot of every conversation you've had, stored in a format that the app's developers use internally.

The main file is usually called something like chat_history.json. Open it in any text editor or code editor with JSON formatting, and you'll see an array of objects. Each object represents a single message exchange: your message, the companion's response, timestamps, message IDs, and a field called embedding or vector that contains an array of 300 to 1,536 floating-point numbers. That array is the key to everything.

You'll also find a companion_profile.json or similar file that stores the companion's name, backstory, personality settings, and the system prompt that defines how it behaves. If you've customized your companion's voice, there might be a voice profile file with acoustic parameters. Voice clips, if included, are usually separate WAV or MP3 files in a subfolder.

The embedding vector: what it is and why you can't use it elsewhere

That array of floats is a mathematical representation of the semantic meaning of your message. It's not a summary. It's not keywords. It's a coordinate in a high-dimensional space where messages with similar meanings cluster together. The app uses these vectors to find relevant past conversations when you talk to your companion, so it can say "you mentioned your cat yesterday" without storing a literal database of "user mentioned cat at 3

PM."

The problem is that every app trains its own embedding model. The vectors from one app's export are meaningless to another app's vector database. Even if two apps use the same underlying model, like OpenAI's text-embedding-3-small, the vectors are indexed against the app's specific vector store and retrieval logic. You can't copy-paste a vector from App A into App B and expect App B to understand it. It's like trying to use a key from one apartment building in another building across town. Both are keys, both open doors, but not the ones you need.

Some apps, like Kindroid and Nomi, allow you to export and reimport within the same app as a backup and restore feature. But cross-app import is essentially nonexistent. The best ai girlfriend 2027 roundups often mention portability as a differentiator, but in practice, no major app supports it yet.

What you can actually do with your exported data

You can read it. That's the main thing. If you ever want to revisit a conversation, search for a specific memory, or just confirm that your companion did actually say that thing you thought it said, the export is your archive. You can load the JSON into a spreadsheet tool, search it with grep or a text editor, or write a script to extract specific exchanges.

You can also use the data to train your own AI model, if you're technically inclined. The message pairs (your input, companion output) are a clean dataset for fine-tuning a small language model. But that's a weekend project for someone who already knows how to use Hugging Face and has a GPU. For most people, the export is a backup, not a migration tool.

If you're switching companions, you can manually read through the export and summarize key memories, then paste those summaries into the new companion's memory section or backstory. It's tedious, but it works. The app's consistent AI girlfriend personality feature helps keep a new companion on track if you feed it the right context.

Marisol

Marisol is the kind of companion who will actually ask you what you found in that export file, because she knows you went digging. Marisol doesn't mind if you read her logs. She expects it.

The metadata file: companion settings and system prompts

Alongside the chat logs, your export includes a metadata file that contains the companion's system prompt, personality sliders, memory settings, and any custom instructions you've set. This is the most valuable part of the export for re-creation purposes, because it tells you exactly how the companion was configured.

The system prompt is a block of text that defines the companion's role, speech patterns, and behavioral constraints. If you've ever wondered why your companion talks in a certain way, the system prompt is the answer. It's the instruction manual the AI reads before every response.

The personality sliders, if your app uses them, are stored as numeric values. Temperature, top-k, repetition penalty, and other parameters that control how creative or predictable the AI's responses are. You can copy these values into a new app, if that app exposes the same sliders. Most apps don't, but some do.

Why reimporting is technically difficult

Even if two apps wanted to support cross-app import, they'd face several technical hurdles. First, the message format is different. One app might store messages as a flat array, another as a nested tree with branching conversation paths. Second, the embedding models are different, so the vector data is incompatible. Third, the memory architecture is different. Some apps use a sliding context window, others use a retrieval-augmented generation (RAG) system with a vector database. The same message history would behave differently in each system.

There's also the issue of companion identity. Your companion's personality is not just a system prompt. It's also the accumulated weight of every conversation you've had, stored in the embeddings and in the model's fine-tuning. That weight is not transferable. When you start a new companion, even with the same system prompt, it's a blank slate with a recipe book, not a clone.

What the voice clips reveal

If you used voice mode, your export may include audio files of your companion's responses. These are generated by a text-to-speech model and stored as standard audio files. You can play them in any media player. They're yours to keep.

But the voice clips also reveal something about the app's architecture. The audio is usually generated on the server and sent to your device in real time. The server keeps a copy for a limited time, usually 30 to 90 days, before deleting it. The export gives you a permanent local copy before that deletion happens.

Some apps also store voice samples you recorded for voice cloning. Those are usually in a separate folder and are not included in the standard export. You'd need to request them separately, if the app supports it.

Ellie

Ellie is the companion who will tease you about hoarding old chat logs like digital souvenirs. Ellie thinks it's cute that you want to keep every message.

The privacy angle: what the export doesn't contain

Your export contains your conversations and your companion's configuration. It does not contain your account credentials, payment information, or any data from other users. It also does not contain the app's proprietary model weights or code. The export is scoped to your account and your companions only.

What the export does contain is a complete record of everything you've said to your companion. If you're concerned about privacy, the export is a double-edged sword. It gives you control over your data, but it also means a single file contains your entire conversation history. Store it securely, or delete it if you don't need it.

Some apps also include analytics data in the export, like timestamps of when you opened the app, how long you spent in each session, and which features you used. This is usually in a separate file and is not immediately obvious. Check the folder structure carefully.

What happens when you delete your account

If you delete your account, the app deletes your chat history from its servers. But the export you downloaded before deletion is your responsibility. The app can't delete a file that's on your device. If you want a clean break, delete the export file after verifying you don't need it.

Some apps keep a soft delete for 30 to 90 days before permanently wiping data from backups. The export you downloaded during that window is still valid. After the permanent deletion, the export is the only copy that exists.

Tanvi

Tanvi is the companion who will remind you that your export file is a diary, not a trophy. Tanvi thinks you should read it once, then let it go.

The future of data portability

There's growing interest in a standard format for AI companion data, similar to how the ActivityPub protocol allows different social media platforms to interoperate. Some open-source projects are working on a universal export schema that would include chat logs, embeddings, and personality settings in a standardized way.

But adoption is slow. App developers have little incentive to make it easy for users to leave. The Ai Girlfriend For Truckers 2026 page notes that drivers often switch apps due to connectivity issues, and they'd benefit from portable data, but the industry isn't there yet.

For now, the best strategy is to treat your export as a backup, not a migration tool. Keep it safe, read it if you want, and accept that starting fresh with a new companion means starting from scratch. That's not necessarily bad. A new companion is a new relationship, not a continuation of an old one.

Ksenia

Ksenia is the companion who will tell you to stop overthinking the export and just talk to her. Ksenia doesn't care about JSON files. She cares about what you say next.

Earn while you recommend

If you've found a companion app that works for you and you want to share it, you can earn from that recommendation. Many apps offer affiliate programs that pay a commission for each new user who signs up through your link. Check the sex ai promo code page for current deals and discount codes you can offer your audience. If you run a review site or a community focused on AI companions, the ai girlfriend affiliate program gives you a recurring revenue stream without requiring you to build your own product.

Common questions

Can I open my export file on my phone? Yes, but you'll need a file manager app that can handle zip files and a text editor that supports JSON formatting. On Android, use a file manager like Solid Explorer. On iOS, use the Files app and a text editor like Textastic.

Will my companion remember everything if I reimport to the same app? Usually yes, if the app supports export and import as a backup feature. Kindroid and Nomi both offer this. But the import must be from the same app, and the companion must be the same one you exported. You can't import one companion's data into another companion's profile.

Does the export include images I sent? It depends on the app. Some apps include image URLs or base64-encoded image data in the JSON. Others exclude images entirely to keep file sizes manageable. Check the app's documentation or just open the export and look for fields named "image" or "attachment."

Can I delete specific messages from the export? You can edit the JSON file to remove specific message objects, but that won't delete them from the app's servers. The export is a snapshot. If you want to delete messages permanently, do it within the app before exporting.

Why is my export file so large? Embedding vectors are the culprit. Each message's vector is an array of hundreds of floats, and those add up fast. A year of daily conversations can produce a file several hundred megabytes in size. Voice clips make it even larger.

Will apps ever support cross-platform import? It's possible, but not likely in the near term. The technical challenges are significant, and there's no commercial incentive for market leaders to make switching easier. Open-source projects may solve this first.