PollyGPT 3 Released, available via API, too. Train your own bots (updated)

4/18 Update

Great news, everyone! PollyGPT V3 is here.  Pro and Plus Graydient users now have access to more powerful 70 billion parameter models like Llama3 and
Mixtral-8x7B. Try the demo by talking to /polly3bot or edit its training and create your own.  You can also now specify a prompt template in the render prompt, just like a recipe.  Use the $prompt variable to mark where the user’s input will go.  Examples are provided in the prompt creation screen.

As a reminder, this feature is an early preview in beta:  We recommend saving a backup when writing your script, as we’ve just launched this feature and the page may reload as bug fixes are deployed.

Along with Polly V3, the programmable mini bots V2 have arrived, with more abilities, memories, and replies.  The version number appears on your saved bots page, as shown below (Charlie)

Your old chat bots are now V1


You can now use Polly (or any Graydient chatbot persona) via API! Results are super fast and are presented as both plain text and semi-structured data when your bot has been instructed to use <command>..</command>-type tags as per the default.

See here for more information: https://app.graydient.ai/api-help/#chat

It is available via API, WebUI, and Telegram:

Pictured: Telegram Experience

What’s new:

1) Chatbots now supports replies and conversational history. You can reply to a message from your bot in Telegram and it will understand that you are asking a followup question. WebUI chat conversations automatically inherit history.

2) The system prompt to support this is quite different. Chatbot personas are now versioned, and you will have to create a new persona (v2) to enjoy this functionality. (I can help delete old personas to free up their names if you want to use the name of your existing bot) Old bots (v1) should still work as before, but replies to their messages won’t maintain context.

3) Bugfix: Chatbot messages for command types other than <say>, <joke>, <suggest>, etc., won’t be treated as errors. The command type used by the bot will be shown in the Telegram message and WebUI.

4) Your chatbot persona can now be given a unique rendering prompt to use for one-click rendering. If supplied, this will override our default system prompt. You can set this up in the chatbot persona editor. This lets you give a different visual vibe to each bot. And as before, if your chatbot suggests keywords that include a concept, the default (or custom) rendering prompt won’t apply, and the chatbot’s suggestion will be used directly.

5) /polly /raw is being phased out in favor of /factsbot, which understands conversational context.

Major changes going on under the hood to understand the system prompt in a structured, data-oriented way. Next version will support a WebUI experience to configure the persona and give examples that doesn’t involve editing the raw system prompt text directly.  Coming up next: Computer vision: Your bots will have image understanding (!)

3/16 Update

The Graydient Platform is expanding into LLMs: Large Language Models.

This is an early preview for Graydient members, as a proof of concept: A train your own PollyGPT Assistant

At the time of this update, there are already 18 Custom Assistants available.

Use cases:

  • Multi-modal chat bots, meaning smart virtual assistants that can respond in natural language as well as understand images and create them
  • Video game NPC characters, create rich dynamic interactions, role playing
  • Business applications: laser-focused chat assistants

Try the demo: Give your bot a unique personality and abilities

Custom Polly Assistants can be made in the browser and conversations and image generation capabilities are also available from custom bots within PirateDiffusion