Mancer: Weaver (alpha)

Mancer: Weaver (alpha) is a text generation model from Mancer, released by the provider on 2023-08-02. Mancer describes it as an attempt to recreate Claude-style verbosity, with the explicit caveat that you should not expect the same level of coherence or memory. It is meant for use in roleplay and narrative situations, so it is positioned as a creative writing model rather than a general purpose assistant.

The model accepts text input and returns text output, so it is a plain language model with no image or audio handling. It has a context window of 8000 tokens and can produce up to 6000 tokens in a single response. Its knowledge cutoff is 2023-06-30, which matters if your prompts depend on events or information from after that date. Within those bounds, the combination of a moderate context window and a large maximum output makes it suited to generating long, continuous passages rather than short answers.

Because the provider frames Weaver around verbosity and narrative, the natural fits are interactive fiction, character driven chat, scene writing, and other roleplay style tasks where a flowing, talkative voice is an asset. The large output ceiling helps when you want a model to carry a scene forward without stopping early, and the 8000 token context gives you room for a system prompt, character description, and several turns of prior dialogue. It is less obviously a fit for tasks that demand tight factual recall or strict instruction following, since the provider itself warns about coherence and memory.

You can call the model through api14's OpenAI-compatible API using the model id "mancer/weaver". The supported request parameters are frequency_penalty, logit_bias, logprobs, max_tokens, min_p, presence_penalty, repetition_penalty, response_format, seed, stop, temperature, top_a, top_k, top_logprobs, and top_p. That set covers the usual sampling controls plus repetition_penalty, min_p, and top_a, which gives you several ways to steer the model away from loops or toward more varied prose. The seed parameter is available if you need reproducible sampling, and logprobs with top_logprobs let you inspect token level confidence.

The main limits to keep in mind are the 8000 token context, the 6000 token output cap, and the 2023-06-30 knowledge cutoff. The provider's own description also sets expectations: this is an alpha model aimed at roleplay and narrative, and it may not hold coherence or memory as well as the style it imitates. If your workload depends on long conversations, recent information, or precise reasoning, a model built for those strengths would likely serve you better. If you want a talkative, scene oriented text generator behind an OpenAI-compatible endpoint, Weaver is a reasonable candidate to test.