Project December

Every conversation starts with a matrix, and you build it from a short form. The intro paragraph sets the scene and the relationship. The example utterance shows the model how the character opens a conversation. Colors and labels decide how the transcript reads, with the AI on one side and you on the other. That last detail matters more than it sounds. A line labeled "Jessica" lands differently from a line labeled "AI assistant", and the interface never lets you forget which is which.

The site ships with a set of ready made personalities, from historical figures to invented characters. The samples page publishes full conversations with them, so you can read how a matrix behaves before spending a credit. Those transcripts are the honest version of a demo: they show the pauses, the repetition, and the moments where the character drifts.

A $5 account starts with 1,000 credits, and credits are spent per exchange rather than per day. A matrix created through the Project December API stays hidden from your terminal menu and costs nothing to create. It burns credits only when you speak to it, and it keeps existing after your balance hits zero. The meter changes how you use the thing. You learn to write shorter prompts and to stop a conversation when it stops giving you anything.

The Project December API is a single server endpoint with a handful of actions. You send your email, an AI name, and a command, then sign the request with an HMAC-SHA1 hash built from your pass phrase and a sequence number. The sequence number blocks replay attacks and moves forward with every valid command. Two optional parameters earn their keep. A wipe command clears the buffer when a conversation goes bad, and force_short cuts generation at the first complete sentence, which helps when the text feeds a speech tool.

Conversations created through the API can be watched from a browser page or pulled as text with the observe_text action. That call leaves the sequence number alone, so you can watch a running conversation while another script keeps talking to it. For anyone building a demo or logging a long session, this is the feature that makes the service usable as plumbing.

When you create a personality, you also choose the engine behind it. The current build offers the G4 engine, and older transcripts on the site were produced with earlier models. Better engines produce better small talk. They do not fix a thin intro paragraph, which is where most of the quality still comes from.