Character.ai Going Bankrupt? A Self-Rescue Guide to Exporting Your Chat History in Six Weeks

Character.ai has filed for Chapter 11 bankruptcy protection, and the official announcement sets the sale window as "late November or early December." For people who've spent two or three years chatting with the same character, this raises a very concrete question: who actually owns those chat logs.

The announcement itself only promises four things: the platform stays running for now, subscriptions keep charging, the privacy policy stays unchanged, and users can export their data or delete their accounts. But the FAQ answer to the direct question "What happens to user data and chats?" never mentions chat history at all, let alone user-created characters.

According to The Next Web, Disney paid $20 million for a non-exclusive perpetual license to specific IP and some legal claims, plus an agreement to temporarily second former Character.ai employees back to keep the platform "running safely for a while." No report says Disney bought the consumer-facing app you use for chatting. That money is a floor price, and other bidders can raise it—and bankruptcy filings show the company's assets and liabilities are in the $50–100 million range. In other words, the floor price might keep the servers alive a bit longer, but that's about it.

What Your Data Means in a Bankruptcy Proceeding

The Actual Scope of the Official Announcement

The four promises sound reassuring, but look closely and you'll notice a key absence: user-created chat history. Someone in the Hacker News thread dug up precedents from 23andMe and Spirit Airlines—after bankruptcy, user data gets sold wholesale as an asset. "The privacy policy stays unchanged" describes today's state; it doesn't tell you what a buyer can do with your logs after the deal closes.

According to official updates, the court-supervised sale is expected to close in late November or early December. If no bidder is willing to keep operating the platform by then, the official wording is that it "will be liquidated."

Legal and Ethical Context

Character.ai settled five teen mental health lawsuits in January, and has since faced additional claims from families and U.S. states. The bankruptcy filing lists assets and liabilities in the $50–100 million range. For any buyer interested in acquiring the platform, these legal burdens and the right to dispose of user data are bundled together.

Chapter 11 bankruptcy protection is a reorganization procedure under Chapter 11 of the U.S. Bankruptcy Code, distinct from Chapter 7 liquidation: it allows a company to keep operating under court supervision while negotiating a debt restructuring plan with creditors, usually aiming to keep the company alive rather than shut it down immediately. After filing Chapter 11, a company enters "debtor-in-possession" (DIP) status—day-to-day operations continue as normal, but major asset dispositions and business-line sales require bankruptcy court approval. This is the legal source of the "court-supervised sale" in this article: whether a sale goes through, who it's sold to, and which assets are sold all have to go through the court process, constrained by creditor committees and bidding rules. Understanding this helps explain why the official announcement only promises "temporary operation," why the sale window is a date range rather than a fixed time, and why bidders above the floor price can jump in.

23andMe is a consumer genetic testing company. After filing for bankruptcy in 2023, its genetic data from millions of users became part of the bankruptcy estate, sparking huge controversy and ultimately being acquired by a nonprofit research institution to limit data flow. Spirit Airlines similarly faced the disposal of customer data and loyalty program assets in its bankruptcy restructuring. These two precedents are frequently cited because they show that user data, while "promised protection" by a privacy policy, can still be transferred or sold wholesale as part of a company's assets—and privacy policies often can't prevent that transfer.

Exporting Chat History Through Official Channels

The Process

This is what I recommend you do first—it only takes two minutes. However, when I tried to open Character.ai's help center, it returned a 403, so the following path comes from third-party guides, and those guides contradict each other. Menu names may have changed.

Web: Log into character.ai, open Profile Settings, then Account, then Manage Account & Data, and choose the data export option.

Mobile: In account settings, look for Export my data (one guide says it's under Settings → Privacy, another says Legal & Privacy). Official updates say both the app and web can export, but one guide claims there's no such option on desktop. If you can't find it on web, use your phone.

Key Limitations

• There's no per-chat export. You get a single archive of your entire account.
• It's a snapshot. New chats after your request aren't included. Request again before you stop using the service.
• Delivery is reportedly by email. Download links expire (one guide guesses 24–48 hours), and arrival time ranges from an hour to a few days. The archive reportedly contains a folder named after your username, with a data/ subfolder holding user.json, character.json, and message.json. I haven't seen Character.ai officially confirm this file list—verify it yourself once you open the archive.
• Don't spam the button. Each request queues separately.

The exported archive is raw JSON—greppable, but not directly importable into other apps.

Exporting Character Cards and Chats with CAI Tools

Why Most People Choose It

To convert data into a format other apps can import, most people turn to CAI Tools. SillyTavern's official documentation points to it for importing Character.ai chats and bots. According to its Firefox store page, it can download chats as "readable" or "transferable" formats and download characters as cards; the Chrome version lists JSON and PNG cards, and PNG cards are exactly what SillyTavern can import.

The Permission List You Must Read Before Installing

Here's what you agree to when installing, based on the public store page:

Item	CAI Tools (Firefox, v3.8.2)
Last updated	October 9, 2026
Users	5,597
Required permissions	Browser tabs, site data for character.ai and chatbotmanager.net
Optional permissions	character.ai, chub.ai, jannyai.com, chatgpt.com, janitorai.com
License	All rights reserved
Paid features	"Some features may require payment" (Patreon)

Its GitHub repo states plainly: "The code here is not the latest." The repo is only for bug reports. You can't audit the build you install using that source.

So you're giving a closed-source extension access to your logged-in Character.ai session, in the same tab where your most private conversations live. I still think it's the right choice, because nothing else exports data into a format another app can load. But proceed carefully:

• Install it in a separate browser profile, logged into Character.ai only.
• Export everything in one go, then uninstall.
• If you're paranoid (reasonably so), log out of all sessions and change your password.
• A flood of similarly named "Character.ai exporters" will appear in stores this month. Don't install any of them unless you can explain who wrote them.

Also, open each exported character card and check the definition field. If it's empty, all you saved is a name, an avatar, and a greeting—nothing else.

PNG character cards are a file format that embeds character definition data in image metadata (tEXt chunks), originating from SillyTavern and its predecessor community conventions. The card looks like an avatar on the surface, but when opened with an app that supports the format, it reads fields like character name, description, personality, dialogue examples, and greeting. If this metadata is missing (say, the export tool only saved the avatar), importing into SillyTavern leaves you with an empty shell character—which is why the article specifically warns you to check the definition field. SillyTavern is an open-source local frontend. It doesn't provide models itself; it acts as an interface connecting to various APIs (Ollama, OpenAI, etc.), managing character cards and chat files—which is exactly why it can accept Character.ai export data.

Rebuilding Characters Locally with SillyTavern and Ollama

The Appeal of Local Deployment

Once your character is a PNG card and your chats are files on disk, you no longer need Character.ai's servers. Local setup takes about ten minutes. According to Ollama's official FAQ, the default context window is 4096 tokens—a character definition plus a long chat history will blow right through it, and the character will start forgetting who it is.

Key Configuration Steps

Install Ollama and pull a model. On Linux:

curl -fsSL https://ollama.com/install.sh | sh
ollama pull llama3.1:8b

Pick a model that fits your RAM or VRAM. Roleplay quality is a matter of personal taste—try several.

Raising the context window matters more than which model you pick. The FAQ's example is a server-wide setting:

OLLAMA_CONTEXT_LENGTH=8192 ollama serve

There's a gotcha on Linux: the install script already runs Ollama as a systemd service, so the above command will throw "address already in use." Put the variable into the service instead:

sudo systemctl edit ollama

**Context length** refers to the total number of tokens a model can "remember" in a single conversation, including system prompt, character definition, history, and current input. Models like Llama 3.1 natively support longer contexts, but Ollama limits the window to 4096 tokens by default for memory and speed reasons. In roleplay scenarios, a detailed character card alone can run thousands of tokens, and adding multi-turn history easily hits the ceiling—the model starts "forgetting" earlier content or even its own character setup. That's why the article calls raising OLLAMA_CONTEXT_LENGTH more important than picking a model. Note that a larger window uses more VRAM/RAM; an 8B model at 16K context typically needs around 8–10GB of VRAM, depending on quantization level.

**systemd** is the service manager for most modern Linux distributions. When you install Ollama via the official script, it's usually already registered as a systemd service running in the background, so running `OLLAMA_CONTEXT_LENGTH=8192 ollama serve` directly in a terminal fails because port 11434 is occupied—the environment variable only affects that one shell launch, not the already-running service. The correct approach is to add an Environment directive to the service unit via `systemctl edit`, then restart, so the config takes effect on every boot.
# Add under [Service]:
#   Environment="OLLAMA_CONTEXT_LENGTH=8192"
sudo systemctl restart ollama

Ollama binds to 127.0.0.1:11434 by default—keep it that way. The FAQ shows how to set OLLAMA_HOST to 0.0.0.0:11434 to expose it to the network—if you do that, anyone on the network can talk to that server.

Then install SillyTavern from the release branch:

git clone https://github.com/SillyTavern/SillyTavern -b release
cd SillyTavern
./start.sh

You'll need git and Node.js installed first. It opens your browser, by default at http://127.0.0.1:8000.

Connect the two: In the API connections panel, select Text Completion, set API type to Ollama, enter http://127.0.0.1:11434, then pick a model from the dropdown.

Import order: Import the PNG card from the character list first, because chats must belong to a character. Then open that character, open Manage chat files from the bottom-left options menu, and use the import button there to select the exported chat file.

After that, your character lives in a folder you can back up with rsync, put into git, or copy to a USB drive.

Export Checklist and Follow-Up Decisions

1. Request the official export now. Request again before you leave.
2. Download the email link the moment it arrives.
3. Use CAI Tools in a separate browser profile to export cards and chats, then uninstall.
4. Check that each card actually contains a definition.
5. Set up SillyTavern + Ollama and import a character to verify the pipeline works.
6. Decide whether to delete your account before the sale closes. You already have your own copy—whether the buyer gets one too is up to you.

A local 8B model isn't as good as what Character.ai ran at its peak. But I'd rather accept that gap than have my chat logs appear on an asset list in a Delaware courtroom.

One-line Summary

Character.ai has filed for bankruptcy, the ownership of user chat logs is unclear, and the article provides a complete plan for exporting data and rebuilding characters locally with SillyTavern.

Paragraph Summary

Character.ai has filed for Chapter 11 bankruptcy protection, with a court-supervised sale expected to close in late November or early December. Disney's $20 million bid only bought a non-exclusive perpetual license to specific IP and some legal claims—the consumer app itself isn't part of the deal. The official announcement promises the platform stays running temporarily, subscriptions keep charging, the privacy policy stays unchanged, and users can export or delete data—yet says nothing about the ownership of chat logs or user-created characters. Precedents from 23andMe and Spirit Airlines show user data can be sold wholesale as an asset. The article advises users to immediately export a JSON snapshot of their entire account through official channels, then use CAI Tools to export SillyTavern-importable PNG character cards (noting the extension is closed-source and requires a logged-in session, so use it in isolation and uninstall afterward), and finally run characters locally via SillyTavern plus Ollama, manually raising Ollama's default 4096-token context window.