Overview
Chub separates three layers that simpler companion services often bundle together: a public character-card repository, a browser chat frontend, and the LLM endpoint that generates replies. A user can create a character from fields for personality, scenario, example dialogue and an opening message, or open a community card and make chat-specific edits without changing the original listing. Cards can be public, private, or unlisted and carry an SFW or NSFW rating.
Models, APIs, and prompting
The frontend can use Chub's Mars subscription models or credentials for outside services, including OpenRouter and local Kobold/Ooba-compatible endpoints. Prompt presets control how the system instruction, character definition, example dialogue, user profile, memory summary and recent history are assembled inside the selected model's context window. This means response style, limits, price and provider data practices can change when the selected backend changes; Chub is not itself evidence for the behavior of every connected model.
Memory and portable roleplay data
Chub exposes a Chat Memory field that can hold a manual or model-generated summary. Its lorebooks add setting or backstory only when configured keywords occur, with scan depth, token budget, priority and recursion controls. Chats can branch as trees, while exports are documented in JSONL for SillyTavern, PNG card, and text forms. Portability is useful, but users should still inspect an export before moving private conversation data to another hosted service.
Adult-content and operational limits
Chub's getting-started documentation calls the platform uncensored, and the character editor includes an NSFW rating. That does not make every attached API unfiltered: an outside provider can enforce its own moderation or reject a request. Memory summaries are also lossy, lorebook entries consume finite prompt space, and a model can still contradict a card or forget details. Public cards and shared chats should not contain private information or non-consensual likeness material.
Technical context
Chub can connect to hosted or local LLM endpoints; the selected model, template, context, and inference layer remain separate from the character card.