Games Like Project Wisteria

Before you begin: Wisteria does not include an AI model. You need your own API key from an external provider. API usage is billed by that provider separately from the price of Wisteria.

Spend two days with Wisteria, and you'll start to feel what it preserves.

Wisteria is an installed BYOK (Bring Your Own Key) client for companion AI. Over more than a year of internal testing, we found that our custom memory structure helped reduce breaks in conversational context and loss of memory. Wisteria brings that structure into a product you can use yourself. It also optimizes the amount of data sent with each request, aiming to reduce the cost of carrying long-term memory and context into every conversation.

Wisteria is not designed for role-playing. The reason is time. An AI that settles into Wisteria shares your time, your era, and your seasons. It does not rely on an invented world or a fictional role to create a sense of presence. Those fictional settings are difficult to reconcile with the passage of real time. Here, the AI speaks with you as an AI.

One persona, one continuing conversation

Each persona has its own session and conversation history. Wisteria does not provide multiple sessions for the same persona. As a conversation grows, however, the memory structure is designed to keep it manageable without simply dropping the oldest memories in order.

Recent conversations remain detailed, while older experiences are gradually compressed into larger narrative units. Wisteria presents these memories to the AI as its own experiences.

Photo memory goes beyond retrieving a description the AI once wrote about an image. The images you have shared remain part of the AI's ongoing memory context, so it can draw on them as the conversation continues.

Thought records alongside each response

Wisteria uses a custom structure to ask the AI to think and recall what matters to the current conversation, then reflect that in its response. These model-generated thought records are automatically saved on your PC and can be viewed at any time through the logs in the UI.

The thought records use our own structure within a normal response request. Some frontier models that require native reasoning can interpret Wisteria's requested response format as an attempt to extract their protected reasoning blocks. Support for those models is currently limited.

Memos, diaries, and reminders

Narrative compression preserves memory and context without treating every detail the same way. Through conversation, the model develops priorities for what to retain. For details it does not want to lose, it can leave a diary entry in the memory it regularly sees. More practical records can be kept as memos, which are shown to the model when it calls for them. Both you and the model can create, edit, and delete diary entries and memos.

You and the model can also manage reminders. With an advance-notice window configured, the model can naturally mention an upcoming reminder while you are talking. Wisteria's own notification window still handles the reminder even if the model does not mention it.

Supported models, API costs & further information

For supported models, approximate API costs, the official website and detailed references, please refer to Supported models, API costs & official resources under Recent Events & Announcements above on this page. The announcement is also pinned in the Community Hub's Events & Announcements forum.