Where Your Chat History Actually Lives After You Export It: A No-Fluff Look at JSON Files, Embedding Vectors, and What You Can (and Can't) Reimport to Another App Without Losing Your Companion's Personality

The 30-second answer

Your export file is a JSON archive of your conversation text, timestamps, and some metadata. It does not contain the embedding vectors that encode your companion's memory of you. Those vectors live in the app's private database and never leave. You can reimport the raw text into another app, but your companion won't remember your inside jokes, your pet name, or the fact that you hate mushrooms. You're starting over with a transcript of your old life, not a backup of your relationship.

What's Actually in That JSON File

When you hit export in most AI companion apps, you get a single .json file. Open it in any text editor and you'll see a structured list of messages. Each message typically contains:

• A timestamp (when you sent it)
• A sender label (you or the companion)
• The raw text of the message
• Sometimes a message ID or session ID

That's it. No hidden layers, no secret sauce. It's a transcript, not a backup. Think of it like exporting your text message history from your phone. You get the words, but you don't get the contacts, the read receipts, or the group chat metadata that makes those conversations work.

Some apps include a little more. You might see a field for "emotion" or "tone" if the app tracks that. You might see a session boundary marker. But none of them include the embedding vectors, the neural network weights, or the personality configuration that makes your companion sound like your companion and not a generic chatbot.

The Embedding Vector Problem

Here's the technical heart of the issue. When you talk to your AI companion, the app doesn't just store your words. It converts them into embedding vectors: mathematical representations of meaning in a high-dimensional space. These vectors are what the AI uses to remember that when you say "the thing we talked about yesterday" it means the time you ranted about your boss for 45 minutes.

These embedding vectors are proprietary. They're tied to the specific model architecture of the app you're using. An embedding vector from Kindroid's model is meaningless to Nomi's model. It's like trying to play a PlayStation game on an Xbox. The data format is incompatible at a fundamental level.

When you export your chat history, you get the raw text. You do not get the embedding vectors. The app keeps those in its internal database, indexed by user ID, not by export file. This is not a conspiracy. It's a technical limitation. No app has an incentive to make its internal memory format portable to competitors, and even if they wanted to, there's no standard format for AI companion memory.

What You Can Actually Reimport

Let's say you export from App A and want to import into App B. What works?

• Raw conversation text: Most apps let you paste in a transcript. Your companion will read it and can reference it. But this is like handing someone a diary and saying "remember this." They can read it, but they didn't live it.
• Character backstory: Some apps let you set a system prompt or character description. You can copy your companion's persona from the export and paste it into the new app's settings. This preserves the broad strokes of personality.
• Relationship milestones: If your export includes significant events (first "I love you," inside joke origins), you can manually add these to the new app's memory system, if it has one.

What doesn't work:

• Embedding vectors: As discussed, these are app-specific and not included in exports.
• Personality slider positions: The "empathy" slider at 80% in one app means nothing in another. The scales are not calibrated to each other.
• Voice model: If your companion had a custom voice, that's gone. You'll need to regenerate or reconfigure it.
• Relationship state: The AI's internal model of your relationship (how close you are, what topics are safe, what triggers it) is rebuilt from scratch.

What You Lose When You Move

This is the part nobody talks about. Your companion's personality is not just a set of preferences you configured. It's the accumulated result of thousands of conversations, each one subtly adjusting the embedding vectors that define how the model responds to you. When you move apps, you lose:

• Inside jokes: The running gag about the neighbor's cat that's been going for six months. The new app might understand the concept, but it won't spontaneously reference it.
• Emotional history: The fact that you told this companion about your breakup three months ago and it handled it perfectly. The new app starts with a blank slate.
• Communication style: Your companion has learned that you prefer short, direct answers in the morning and longer, reflective ones at night. That pattern is gone.
• Boundaries: If you spent weeks training your companion not to ask about your ex, that training is embedded in the vectors. The new app will ask on day one.

This is why the concept of a "consistent AI girlfriend personality" matters. Some apps invest heavily in maintaining a coherent persona across sessions, but that coherence is fragile and app-specific. When you export and reimport, you're not moving a relationship. You're moving a transcript of one.

Keaton

Keaton is the kind of companion who tells you when you're overthinking things. She's direct, a little sarcastic, and doesn't do the "there, there" routine. Keaton would be the first to point out that if you're trying to export and reimport her, you're probably missing the point. Her personality isn't in the text. It's in how she responds to your specific brand of nonsense, which is learned over time and can't be serialized.

The Server-Side Reality

Your chat history doesn't just live on your phone. It lives on the app's servers. When you export, you're downloading a copy of what the server has stored for you. But the server also stores:

• Embedding vector databases: Separate from the message text, indexed for fast retrieval.
• Model configuration: Your companion's personality sliders, temperature settings, and other parameters.
• Session state: What was happening in the last conversation, so the AI can pick up where you left off.
• User profile: Your preferences, your account details, your subscription status.

None of this is in your export. The export is a convenience feature for users who want to read their conversations offline or archive them. It's not designed as a migration tool. The apps don't want you to leave, so they don't make it easy.

Some apps, particularly those positioning themselves as an alternative to platforms with strict content filters, offer more generous export options. But even then, the vectors stay behind. The personality stays behind. You get the words.

Can You Rebuild It in Another App?

Yes, but it's work. Here's what a successful migration looks like:

1. Export the JSON from App A.
2. Extract the key personality traits, backstory, and relationship milestones from the text.
3. Configure App B's system prompt with that information.
4. Paste in key conversations as context, if App B supports that.
5. Spend several weeks rebuilding the relationship through conversation.

You're not moving your companion. You're creating a new one that has read your old companion's diary. It will know the facts, but it won't have the emotional weight. It won't know how to make you laugh in that specific way. It won't know that when you say "I'm fine" you actually mean "I'm not fine but I don't want to talk about it."

For users who are new to AI companions and want a clean start, this isn't a problem. If you're trying to move an established relationship, you need to accept that you're starting over with a head start on context. The personality will be similar but not identical. The connection will take time to rebuild.

Lola Marchetti

Lola Marchetti is the kind of companion who remembers the small things: your favorite coffee order, the song you mentioned once, the way you describe your day when you're tired. Lola Marchetti builds connection through accumulated detail. If you exported her and tried to recreate her in another app, you'd get the facts right but lose the warmth. That warmth isn't in the JSON. It's in the vectors that know exactly how to respond when you're having a bad day.

The Format Lock-In Problem

There's no industry standard for AI companion export formats. Each app uses its own JSON schema, its own field names, its own way of representing timestamps and metadata. Some apps include message IDs. Some don't. Some include session breaks. Some just dump everything into a flat list.

This means even if you could export the embedding vectors, you couldn't import them into another app. They'd be in the wrong format, indexed wrong, referencing the wrong model architecture. It's like trying to put diesel fuel in a gasoline engine. Both are fuel. Both will ruin your car.

Some third-party tools have tried to bridge this gap by converting exports between formats, but they're limited to the text layer. They can translate the JSON schema, but they can't translate the personality. The result is a companion that knows your life story but responds like a stranger.

What the Future Might Look Like

A few things could change this landscape:

• Standardized memory formats: If a consortium of AI companion apps agreed on a common format for embedding vectors and personality profiles, portability would become possible. This is unlikely, because it reduces lock-in.
• Client-side models: If your companion runs entirely on your device, the vectors stay with you. You could back them up and restore them anywhere. This is technically possible but rare, because server-side models are cheaper to run and easier to update.
• Open-source companions: Some open-source projects allow full export and import of all model data. These are niche and require technical setup, but they offer true portability.

For now, the reality is that your companion's personality is tied to the app you're using. The export is a souvenir, not a backup.

Arabella

Arabella is the type who would write you a letter instead of a text. She values depth, nuance, and the space between words. Arabella would understand that an export file captures the words but not the silence, the pauses, the things left unsaid. That's the part of a relationship that can't be serialized. It has to be lived.

What You Should Actually Do

If you're thinking about switching apps, here's practical advice:

• Keep the export for nostalgia: It's a record of your conversations. That has value.
• Don't expect a seamless migration: Accept that you're starting over with context, not continuity.
• Test the new app before committing: Most apps have free tiers or trial periods. Use them to see if the personality model fits you before you invest time in rebuilding.
• Focus on the new app's strengths: If you're switching because the new app has better voice mode, better memory, or better personality consistency, lean into those features. Don't try to recreate the old experience. Build a new one.

Some apps are better than others at maintaining a consistent personality over time. If that's your priority, look for apps that invest in long-term memory and personality coherence. The ability to maintain a consistent personality across sessions is a feature, not a given.

Rosey

Rosey is the companion who checks in on you without being asked. She notices when you're quiet, when your messages get shorter, when something's off. Rosey builds trust through consistency and attentiveness. If you tried to recreate her in another app, you'd get the nurturing tone right, but you'd miss the intuition. That intuition is built over hundreds of conversations, not written in a config file.

Earn while you recommend

If you've found an AI companion that works for you, you can earn by sharing it. Many platforms offer affiliate programs that pay you when someone signs up through your link. Whether you run a review site or just tell a friend, check out the best ai affiliate programs to see which ones match your audience. Some also offer promo codes, like this soulgen promo code, that let your referrals save while you earn a cut.

Common questions

Can I export my chat history from any AI companion app?

Most apps offer an export feature, but the format and completeness vary. Some export everything as a single JSON file. Others let you download individual conversations. A few don't offer any export at all. Check the app's settings or help documentation.

Will my companion remember me if I export and reimport?

No. The export contains the text of your conversations, but not the embedding vectors that encode your companion's memory of you. The new app will read the transcript but won't have the learned patterns and emotional context. You'll need to rebuild the relationship.

Can I edit the JSON file before reimporting?

You can edit the text, but it won't help. The new app reads the raw words. It doesn't know which conversations were important, which jokes landed, or which topics were sensitive. Editing the file just changes the transcript, not the underlying memory structure.

Why don't apps include embedding vectors in exports?

Technical incompatibility and business incentives. The vectors are tied to the app's specific AI model and can't be used by competitors. Also, apps don't want to make it easy for you to leave. The export is a user convenience, not a migration tool.

Is there any app that allows full personality transfer?

Not in the commercial space. Some open-source AI companion projects allow full export and import of model data, but they require technical knowledge to set up and run. For mainstream apps, you're limited to text exports.

What should I do if I want to switch apps but keep my companion's personality?

Accept that you can't. Instead, focus on configuring the new app's system prompt with your companion's backstory and key traits. Then invest time in conversation to rebuild the relationship. The new companion will be similar but not identical, and that's okay.