This document covers how to use the bot in a room.
The 🌟 Features page also includes details about how each feature works and can be configured.
This is related to the 💬 Text Generation feature.
If there's a text-generation handler agent configured, the bot may respond to messages sent in the room.
See screenshots of:
- 🖼️ the default Text Generation flow in 1:1 rooms
- 🖼️ the Text Generation flow in multi-user rooms (where the 🗟 Prefix Requirement setting is auto-configured to "required")
- the on-demand involvement feature
Whether the bot responds depends on:
-
🛠️ configuration whether there's a configured
text-generationhandler agent (or acatch-allhandler agent). See Mixing & matching models -
(🎨 agent capabilities) whether the configured
text-generation(orcatch-all) handler agent actually supports text-generation. The provider may lack support for this feature or it may be disabled in the 🤖 agents configuration -
(the 🗟 Prefix Requirement setting) whether a prefix (e.g.
!bai) or user mention (e.g.@baibot) is required for messages sent to the room. For multi-user rooms, this setting defaults to "required". See 🌟 Features / 💬 Text Generation / On-demand involvement for details.
Room messages start a threaded conversation where you can continue back-and-forth communication with the bot. Using on-demand involvement, you can can also mention the bot to provoke it to get involved in any conversation thread or reply chain.
Unless you've enabled the ♻️ Context Management feature, all messages will be sent to the agent's API each time. If the context management feature is enabled, older messages may be dropped.
This is related to the 🗣️ Text-to-Speech feature.
If there's a text-to-speech handler agent configured, the bot may convert text messages sent to the room to audio (voice).
See:
-
a 🖼️ screenshot of the bot's Text-to-Speech-only mode
-
a 🖼️ screenshot of the bot's Seamless voice interaction mode
By default, the bot:
-
will offer tex-to-speech for its own messages which are a response to voice message from your, as part of the Seamless voice interaction feature. This can be adjusted via the 🗣️ Text-to-Speech / 🪄 Bot Messages Flow Type setting.
-
does not turn your own text messages to audio (voice). If you'd like for the bot to operate in such a mode, use the 🗣️ Text-to-Speech / 🪄 User Messages Flow Type setting (see Text-to-Speech-only mode).
This is related to the 🦻 Speech-to-Text feature.
If there's a speech-to-text handler agent configured, the bot may transcribe voice messages sent to the room to text.
See a 🖼️ Screenshot of the default flow for Speech-to-Text and Text-Generation.
The speech-to-text feature triggers automatically by default, but can be adjusted via the 🦻 Speech-to-Text / 🪄 Flow Type setting.
If all your messages are in the same language, you can improve accuracy & latency by configuring the language (see 🦻 Speech-to-Text / 🔤 Language).
This is related to the 🖌️ Image Generation feature.
This feature is not configurable at the moment. The configuration (size, quality, style) specified at the 🤖 agent level will be used.
Simply send a command like !bai image A beautiful sunset over the ocean and the bot will start a threaded conversation and post an image based on your prompt.
See a 🖼️ Screenshot of the Image Generation feature.
You can then, respond in the same message thread with:
- more messages, to add more criteria to your prompt.
- a message saying
again, to generate one more image with the current prompt.
A variation of generating images is to generate "sticker images".
See a 🖼️ Screenshot of the Sticker Generation feature.
To generate a sticker, send a command like !bai sticker A huge ramen bowl with lots of chashu and a mountain of beansprouts on top.
The difference from generating images is that the bot will:
- generate a smaller-resolution image (currently hardcoded to
256x256) - smaller/quicker, but still good enough for a sticker - potentially switch to a different (cheaper or otherwise more suitable) model, if available
- post the image directly to the room (as a reply to your message), without starting a threaded conversation
Some models (like OpenAI's Dall-E-3) can only generate larger images (1024x1024, etc., for a higher charge), so we switching to a smaller/cheaper model (like Dall-E-2) is a way to generate a sticker cheaply.