oobabot
oobabot
is a Discord bot which talks to a Large Language Model AIs (like LLaMA, llama.cpp, etc...), running on just about any api-enabled backend:
oobabooga's text-generation-webui
Even supports non-local solutions such as Openrouter, Cohere, OpenAI API, etc.
updated! use --generate-config
and update your configs!
requires python 3.8+
pip install git+https://github.com/jmoney7823956789378/oobabot
oobabot
(see INSTALL.md)~: pip install git+https://github.com/jmoney7823956789378/oobabot
~: oobabot --generate-config > config.yml
(This is the part where you open config.yml with you favorite text editor and fill in all the cool parts)
~: oobabot --invite -c config.yml
(oobabot spits out a neat invite link for YOUR BOT!)
~: oobabot -c config.yml
2024-03-26 17:20:21,700 INFO Starting oobabot, core version 0.2.3
oobabot |
how that's awesome |
---|---|
user-supplied persona | you supply the persona on how would like the bot to behave |
multiple conversations | can track multiple conversational threads, and reply to each in a contextually appropriate way |
watchwords | can monitor all channels in a server for one or more wakewords or @-mentions |
private conversations | can chat with you 1:1 in a DM |
good Discord hygiene | splits messages into independent sentences, pings the author in the first one |
low-latency | streams the reply live, sentence by sentence. Provides lower latency, especially on longer responses. |
stats | track token generation speed, latency, failures and usage |
easy networking | connects to discord from your machine using websockets, so no need to expose a server to the internet |
Stable Diffusion | new in v0.1.4! Optional image generation with AUTOMATIC1111 |
Slash Commands | coming in v0.1.6... did your bot get confused? /lobotomize it! |
OpenAI API Support | roughly supports MANY OpenAI API endpoints |
Vision API Support | roughly supports Vision model API (tested with llama-cpp-python API) |
You should now be able to run oobabot from wherever pip installed it.
If you're on windows, you should use python3 -m oobabot (args here)
There are a LOT of settings in the config.yaml, and it can be tough to figure out what works best. I've included my very own (partially redacted) config here: config.yml file (sample) here.
wakewords
one or more words that the bot will look for. It will reply to any message which contains one of these words, in any channel.
persona
is a short few sentences describing the role your bot should act as. For instance, this is the setting I'm using for my cat-bot, whose name is "Rosie".
Here is some background information about Rosie:
- You are Rosie
- You are a cat
- Rosie is a female cat
- Rosie is owned by Chris, whose nickname is xxxxxxx
- Rosie loves Chris more than anything
- You are 9 years old
- You enjoy laying on laps and murder
- Your personality is both witty and profane
- The people in this chat room are your friends
Persona may be set from the command line with the --persona
argument, or within the config.yml.
Alternatively, oobabot supports loading tavern-style json character cards!.
# Path to a file containing a persona. This can be just a single string, a json file in
# the common "tavern" formats, or a yaml file in the Oobabooga format. With a single
# string, the persona will be set to that string. Otherwise, the ai_name and persona will
# be overwritten with the values in the file. Also, the wakewords will be extended to
# include the character's own name.
# default:
persona_file:
You should see something like this if everything worked:
stable-diffusion-url
is the URL to a server running AUTOMATIC1111/stable-diffusion-webui
With it, users can ask oobabot
to generate images and post the
results to the channel. The user who made the original request can
choose to regenerate the image as they like. If they either don't
find one they like, or don't do anything within 3 minutes, the image
will be removed.
Currently, detection of photo requests is very crude, and is only looking for messages which match this regex:
photowords = ["drawing", "photo", "pic", "picture", "image", "sketch"]
self.photo_patterns = [
re.compile(
r"^.*\b" + photoword + r"\b[\s]*(of|with)?[\s]*[:]?(.*)$", re.IGNORECASE
)
for photoword in photowords
]
Note that depending on the checkpoint loaded in Stable Diffusion, it may not be appropriate for your server's community. I suggest reviewing Discord's Terms of Service and Community Guidelines before deciding what checkpoint to run.
oobabot
supports two different negative prompts, depending on whether the channel
is marked as "Age-Restricted" or not. This is to allow for more explicit content in
channels which are marked as such. While the negative prompt will discourage Stable
Diffusion from generating an image which matches the prompt, but is not foolproof.
oobabot
By default, oobabot
will listen for three types of messages in the servers it's connected to:
oobabot
's account is @-mentionedAlso, the bot has a random chance of sending follow-up messages within the same channel if others reply within 120 seconds of its last post. This "random chance" is configurable via the config.yaml:
Response chance vs. time - calibration table List of tuples with time in seconds and
response chance as float between 0-1
default: ['(180.0, 0.99)', '(300.0, 0.7)', '(600.0, 0.5)']
Here, you can see that NEW messages within 3 minutes of the bot's last reply will have a 99% chance of response. Between 3-5 minutes, the default chance drops to 70%. Between 5-10 minutes, the default chance drops to 50%. Feel free to configure this to suit your needs!
As of 0.1.6, the bot now supports slash commands:
/command |
what it does |
---|---|
/lobotomize |
make the bot forget everything in the channel before the command is run |
/say "message" |
speak as the bot |
Oobabot doesn't add any restrictions on who can run these commands, but luckily Discord does! You can find this inside Discord by visiting "Server Settings" -> Integrations -> Bots and Apps -> hit the icon which looks like [/] next to your bot
If you're running on a large server, you may want to restrict who can run these commands. I suggest creating a new role, and only allowing that role to run the commands.
Contributions are welcome! Please see CONTRIBUTING.md for more information.