OhChat: The Longer Answers

The headline score is a plain average of five sub-scores, each marked out of ten. OhChat took a ten for memory and continuity, nines for conversation quality and character consistency, and eights for media features and value. Averaged, that lands just under nine. We publish the sub-scores precisely so the headline number can be argued with rather than accepted.

Yes, when the platform changes. Model updates, new media features and changes to the free allowance all move sub-scores, and the review carries the date of the last full test cycle. We do not adjust the number based on user sentiment or on how the app is being discussed elsewhere. A revision means the test sequence was run again from a fresh account.

Both, in that order. The conversation and memory tests run on a free account, because that is what a new user actually gets and because the same model answers either way. Media features and the value sub-score require a paid tier, so those are tested afterwards on an upgraded account. The two phases are kept separate so a paywall cannot flatter the conversation result.

Better than at the start, with one caveat. Accumulated context makes replies more specific, and the character stops asking things it has already been told. The caveat is thread length: past several thousand messages, the oldest specifics begin to fade and the character starts filling gaps from its base profile. Moving the details that matter into the character description prevents that drift entirely.

You can keep as many separate threads as you like, and memory does not bleed between them. Each character maintains its own profile and its own history. What does not work well is putting several companions into one scene together: coherence breaks down within a handful of turns, and the characters begin answering for each other. Treat OhChat as a set of one-to-one conversations.

It usually accepts the correction and carries the new version forward. In our testing a single clear statement was enough, and the old detail did not resurface later in the same session. Ambiguous corrections work less well, because the model may keep both versions and choose between them unpredictably. State the replacement plainly rather than hinting that something was wrong.

Threads can be deleted individually, and deleting one also removes the memory attached to it, so the character reverts to its base description. Characters you built yourself can be removed separately. Deleting the account removes the login, the threads and the custom characters together. Anything the payment processor holds for its own records sits outside the chat product and is governed by its terms.

The age gate runs before any chat interface loads, and the confirmation is required again on a new device or after clearing site data. It is a self-declaration, not identity verification, so it is a boundary rather than a wall. Every character on the platform is fictional and generated, and requests aimed at real individuals or anyone underage are refused by the model.

Paid tiers run as recurring subscriptions handled by the payment processor, and cancelling is done from the account settings rather than by email. Cancelling stops the renewal; access continues to the end of the period already paid for. Your threads and characters remain in the account afterwards, reverting to free-tier message limits rather than being deleted.