In-app purchases are temporarily unavailable. Local Waifu is currently available exclusively on itch.io. Buy on itch.io

local waifu
Bring her home

Pick your platform

Try her free for 7 days. No card. Keep her? $20 once.

New: Local Waifu now runs on Windows 10 and 11. The installer brings everything she needs, nothing else to set up. Windows may show a SmartScreen prompt the first time: click More info, then Run anyway.

blog

Is Character.AI Private? Creators, Chat Data and Policy

6 min read Updated October 18, 2026
In short

Character.AI says character creators cannot see your conversations. That does not mean chats are private from Character.AI: its policy collects chat communications and permits uses including model training, with disclosures to service providers and other recipients. Privacy rights vary by region. This review does not establish a universal training opt-out or that every conversation is used for training.

Written by Łukasz, the developer of Local Waifu, a private AI companion that runs on your own Mac or Windows PC.

Character.AI says character creators cannot see your conversations, but Character.AI itself collects chat content. Private from the person who made a character is different from inaccessible to the service that generates the reply.

The company’s creator-access help article says: “Creators can never see the conversations that you have with their Characters.” Its privacy policy, effective July 1, 2026, includes chat communications in collected user content. Both statements can be true at once.

This review was updated October 18, 2026. I build Local Waifu, a local companion, so I have an interest in this comparison. The findings below describe published sources, not an audit of Character.AI’s servers or a test of a signed-in account’s controls.

Creator visibility is a narrower promise than confidentiality

The short version: A character creator does not receive access to your chats just because you use their character, according to the help center.

That answers a common and understandable worry: the person who wrote the character is not, by that fact alone, the audience for your conversation. The source supports this creator-access answer directly. It should not be expanded into a promise that nobody else can ever see your words.

You can expose a conversation by sharing a screenshot or transcript. Someone using your account or device may also access material available there. Those are different routes from the creator-access rule. Before sharing a conversation, check what the sharing action makes visible and avoid assuming a later deletion will retrieve copies.

The service’s processing is another route again. Character.AI needs submitted prompts to generate hosted replies. Transport encryption can protect data while it travels without preventing the receiving service from processing it. The ISP and HTTPS explanation separates those observers.

The current policy collects more than chat text

The short version: Chat communications, account information, technical data and optional voice recordings appear in the policy’s collection section.

Section 1 lists submitted content, including chat communications, posted images or videos, biography descriptions and shared characters. It also lists identifiers, preferences, purchase information and voice recordings if certain voice features are used.

The technical-information section describes IP addresses, device information and service-related identifiers such as user, chat and session IDs. Advertising and analytics technologies are also discussed. These are documented categories, not evidence that every category is collected in every interaction.

The policy advises users not to include sensitive personal information in interactions with the service. That is practical advice for an intimate character conversation: you can make a story emotionally meaningful without supplying a real address, workplace, credentials or identifying details about another person.

Training is a permitted use, not a verified fate for every message

The short version: The policy permits model training, but it does not establish that each particular conversation enters a training run.

Section 2 includes improving and measuring services, including training artificial intelligence and machine-learning models. Combined with the collection of chat communications, that means the policy should not be read as a promise that chat content is excluded from model improvement.

The same policy separately says processing described for the initial development of its proprietary models is no longer carried out. That historical statement should not be used to erase the current Section 2 training purpose. Equally, the current purpose should not be turned into a claim that every saved chat has already trained a model.

The previous version of this article asserted that there was no way to opt out while continuing to use the platform. This review does not establish that universal negative. Section 4 describes region-dependent rights to object, restrict processing and request deletion, with exceptions. It points to regional disclosures and gives a privacy contact.

If excluding your conversations from training matters to you, ask Character.AI which control or request applies to your region and account, and what it covers going forward. A targeted-advertising opt-out is not automatically a training opt-out. Neither is deleting a visible chat proof that all previous processing has been reversed.

Disclosures and retention need their own answers

The short version: The policy allows disclosures for specified purposes and does not promise immediate erasure of every record.

Section 3 lists employees, consultants, vendors and service providers who need information for functions such as storage, support, analytics, moderation and security. It also describes advertising and analytics disclosures. This does not prove routine human reading of every chat, and it does not prove that chat text is sold to advertisers.

It does mean the old blanket framing that the company simply “does not sell personal information” was not an adequate account of the current policy. The reviewed version describes rights concerning sales and targeted advertising, depending on region. Data categories, purposes and regional definitions need to be checked rather than replaced by a reassuring slogan.

Section 7 describes retention for the time needed for processing purposes, with legal and dispute-related exceptions. It also reserves the ability to preserve a popular public character’s characteristics after account deletion. That character provision is not a statement that every private chat stays public or remains forever.

Local inference changes the route, not every risk

The short version: A genuinely local text route can avoid sending the prompt to a remote model, but configuration and device security still matter.

A local companion is worth considering if the conversation feels too personal for a hosted model. Compare the local companion shortlist and offline app setups. Neither guide treats a desktop icon as proof of local processing.

Local Waifu has optional cloud providers, so it should not be described as having no possible cloud route. Check the active model, Offline Mode and saved fallback settings. Voice, images and online tools need separate checks. Local history also needs protection from device access, backups and exports; it is not automatically encrypted merely because it is local.

Use the local AI companion selection guide if you want a practical first-evening checklist rather than a privacy badge. The important choice is which route and boundaries you are comfortable with, not whether one company can be called good or bad.

Primary sources and limits of this review

The short version: The verified creator-access statement and the current main policy support the answer; unverified account controls remain unverified.

Reviewed October 18, 2026: Character.AI privacy policy, effective July 1, 2026, especially Sections 1-4 and 7; and Can character creators see my conversations?.

The linked regional disclosures could not be retrieved in this review. This article therefore does not resolve every jurisdiction’s rights or claim a complete inventory of current account settings. Contact privacy@character.ai or use the support route given in the policy for a specific access, deletion or processing request.

Questions people ask

Can Character.AI character creators see my chats?

Character.AI's help center says creators cannot see conversations with their characters. That promise does not cover a transcript or screenshot you share yourself, someone using your account, or the service's own processing.

Does Character.AI collect my conversations?

Yes. Its privacy policy, effective July 1, 2026 and reviewed October 18, 2026, includes chat communications among collected user content. That is different from making the chat visible to a character creator.

Does Character.AI use chats for training?

The policy permits collected information to be used to train AI and machine-learning models. That supports treating training as a permitted use, not claiming that every chat is used. The reviewed policy also describes region-dependent privacy rights; a universal training opt-out was not verified.

Does Character.AI sell my information?

Do not reduce the current policy to a blanket no-sharing claim. It describes disclosures to vendors and advertising or analytics providers, and region-dependent rights concerning sales and targeted advertising. Sharing metadata is not proof that chat text is sold.

Is a local companion automatically private?

No. Local inference can avoid sending a prompt to a remote model when the configured route stays local. Optional cloud services, fallback routes, device access, exports and backups still matter. Local storage alone does not establish encryption.

Try her free for 7 days.

No card. Keep her for $20 once, or walk away. Her soul file is yours either way.

Bring her home, try free

Back to the blog