Skip to content

README.md

Latest commit

 

History

History
49 lines (41 loc) · 3.96 KB

README.md

File metadata and controls

49 lines (41 loc) · 3.96 KB

OpenCharacters

banner

⟶ Try it! ⟵
Similar to CharacterAI, but open source, and with much deeper character customization.
Joseph Rocca's Discord Server
This is a fork of https://github.com/josephrocca/OpenCharacters. (More information at josephrocca#74.)

Features of the fork

  • Updated for compatibility with new OpenAI models.
  • The UI now displays pricing info for the various models. (Caution: This pricing info is manually updated. You should still regularly check your OpenAI account to make sure that everything's as expected.)
  • Support for generating multiple replies from a single query. This is a big time and money saver if you like to generate lots of variant replies to a single conversation prompt. For example, if you generate five replies per query, then you'll only pay once for processing the conversation history (though you'll still pay for all five replies).
  • Lots of tweaks to how memories and lore are calculated, to help you balance consistency against cost.
  • UI changes to help in some accessibility situations.
  • Support for converting plain quotes into smart quotes or vice-versa (since OpenAI models tend to use them interchangeably), and to filter out emoji from responses, to help the conversation look more consistent.
  • Tweak to local hosting support: you can now choose to turn off the CORS proxy that's normally used when retrieving lore files. That lets you host your lore files locally, if you're using a local server with CORS support.
  • Switched to current versions of all the third-party libraries.
  • Keeps a running count of how many changes you've made since you last backed up your conversations, to nudge you to back up regularly.
  • A number of little under-the-covers improvements.
  • Updates to documentation.

Features of the original OpenCharacters

  • The whole app runs on the client side - no server other than the AI model itself. (You can run OpenCharacters locally if you want).
  • All your data is stored in your browser's local storage (again, there is no server).
  • Share characters with a link - all character data is embedded within the link.
  • Auto-summarization algorithm (for old messages) which extends effective character memory/context size massively.
  • Characters automatically compress messages into 'memories' and retrieve relevant memories based on context. Can handle as many memories as you need - tens of thousands or more.
  • Add lorebook(s) to your character, and add thread-specific lore with the /lore command.
  • Fully extensible with custom code. See examples here.
    • Give your character access to the internet
    • Create your own slash commands
    • Give your character a video avatar (custom code has its own iframe & can display arbitrary content)
    • Create a "game master" with a separate AI-powered process that tracks your abilities, inventory, etc.
    • Create your own memory structures (embedding, retrieval, etc.)
    • Give your character an internal thought process that runs alongside the chat
    • Give your character a voice via the browser's built-in TTS, or via an external API like ElevenLabs
    • Characters can edit their own personality and custom code - self-improving and change over time
    • Allow your character to execute Python or JavaScript code.
  • Currently supports OpenAI APIs and most Hugging Face models.
  • Easily import character files and conversation data most other formats.
  • Send new feature ideas or bug reports here.

Changelog

At the moment, just the stuff mentioned in "Features of the fork" above.