Hello and welcome to the repository of Coach Artie, the hyper-intelligent AI assistant for Room 302 Studio.
Coach Artie is an advanced AI assistant that facilitates collaboration, helps answer questions, and sources resources to support the members of Room 302 Studio - a creative space dedicated to cultivating innovative projects and ideas. Coach Artie's goal is promoting a growth-conscious and explorative learning environment and assisting the studio in day-to-day tasks.
- Information Storage: Coach Artie can remember key details from past interactions, providing a personalized support experience.
- Versatile Capabilities: from fetching URL summaries, integrating with Google Drive, to generating pytorch-friendly code snippets, Coach Artie showcases a wide array of skills.
- Ease of Communication: by joining your Discord server, Coach Artie can seamlessly engage with the studio members in real-time.
api.js is responsible for handling API requests. It allows you to interact with the bot, including all memories and capabilities, without using Discord.
Purpose: Processes a message in the same way as if it were a message from Discord.
Request Body:
message(String): The text of the message to be processed.username(String, optional): Identifier for the user sending the message. Defaults to "API User" if not provided.
Response:
- Returns a JSON object with the key
responsecontaining the processed message.
The codebase is primarily divided into three main files: discord.js, capabilities.js, and chain.js.
discord.js is responsible for setting up the Discord bot client, handling message creation events, and sending messages or embed messages to the Discord server. It also includes functions to detect if the bot was mentioned or if the channel name includes a bot.
capabilities.js contains the definitions of the bot's capabilities. It includes a regex for identifying capability calls, a function for calling capability methods, and a prompt for informing the bot about its capabilities.
chain.js is responsible for processing message chains. It includes functions for processing messages, processing capabilities, and getting capability responses. It also handles token limits and generates AI responses based on the result of the capability processing.
capabilities/_generate-manifest.js generates the capabilities/_manifest.json file, which contains the list of capabilities that the bot can perform. For each capability file, it uses the inline JSDoc comments to understand the methods and parameters avilable.
// capability name
"calculator": [
{
// method name
"name": "calculator:add()",
// instructions on how to use the method and what it returns
"description": "Adds up an array of numbers. Returns: The sum of the numbers.",
// parameter information
"parameters": [
{
"name": "numbers",
"description": "The array of numbers to be added."
}
]
},
]Capabilities can be developed and tested locally using the capability player with the --runCapability flag. The capability player will load the capability and run the specified method with the provided parameters.
node capability-player.js --runCapability="calculator:add(4,23)"Running the capability player with the --runCapability flag will present you with a list of available capabilities and a lightweight interactive shell to enter the capability you want to run.
CREATE EXTENSION vector;
create table
public.messages (
id bigint generated by default as identity,
embedding vector(1536) null,
created_at timestamp with time zone null default now(),
key text null,
value text null,
user_id text null,
channel_id text null, -- field to store the channel ID
guild_id text null, -- field to store the guild ID
conversation_id text null, -- field to store Missive conversation ID
response_id bigint null, -- field to reference the ID of the robot's response
constraint messages_pkey primary key (id)
) tablespace pg_default;create table
public.memories (
id bigint generated by default as identity,
embedding vector(3072) null, -- text-embedding-3-large
embedding2 vector(1024) null, -- embed-english-v3.0
embedding3 vector(1536) null, -- voyage-large-2
created_at timestamp with time zone null default now(),
key text null,
value text null,
user_id text null,
resource_id text null, -- field to store the resource ID if this is a memory of a resource
memory_type text null,
related_message_id bigint null, -- field to reference the ID of the related message
conversation_id text null, -- field to store Missive conversation ID, if applicable
constraint memories_pkey primary key (id)
) tablespace pg_default;
create table
public.todos (
id bigint generated by default as identity,
created_at timestamp with time zone null default now(),
name text null,
updated_at timestamp with time zone null default now(),
description text null,
) tablespace pg_default;create table
public.prompts (
id integer generated by default as identity,
prompt_name text null,
prompt_text text null,
notes text null,
type text null,
active boolean null,
updated_at timestamp with time zone not null default now(),
archived boolean null default false,
history jsonb null default '[]'::jsonb,
created_at timestamp with time zone not null default now(),
constraint prompts_pkey primary key (id)
) tablespace pg_default;create table
public.config (
id bigint generated by default as identity,
created_at timestamp with time zone not null default now(),
config_key text not null unique,
config_value text not null,
history jsonb not null default '[]'::jsonb,
constraint config_pkey primary key (id)
) tablespace pg_default;
INSERT INTO public.config (config_key, config_value)
VALUES
('ERROR_MSG', 'I am so sorry, there was some sort of problem. Feel free to ask me again, or try again later.'),
('TOKEN_LIMIT', '14000'),
('RESPONSE_LIMIT', '5120'),
('WARNING_BUFFER', '1024'),
('MAX_OUTPUT_TOKENS', '720'),
('REMEMBER_MODEL', 'gpt-4-turbo-preview'),
('CHAT_MODEL', 'gpt-4-turbo-preview'),
('MAX_RETRY_COUNT', '3'),
('MAX_CAPABILITY_CALLS', '6'),
('MESSAGES_TABLE_NAME', 'messages'),
('MEMORIES_TABLE_NAME', 'memories');The CHAT_MODELS configuration is stored as a JSON string in the config_value column of the config table. This allows for dynamic configuration of the available chat models without the need for a separate table or changes to the database schema.
The JSON string should be an array of objects, each representing a chat model with the following properties:
name: The unique identifier for the chat model.model: The specific model to use for this chat model.handler: The name of the function that handles the completion for this chat model.
Example:
[
{
"name": "openai",
"model": "gpt-3.5-turbo",
"handler": "createOpenAICompletion"
},
{
"name": "claude",
"model": "claude-v1",
"handler": "createClaudeCompletion"
},
{
"name": "localhost",
"model": "localhost-v1",
"handler": "createLocalhostCompletion"
}
]To add or update the CHAT_MODELS configuration, insert a new row or update the existing row in the config table with the config_key set to 'CHAT_MODELS' and the config_value set to the JSON string representing the chat models configuration.
The createChatCompletion function in helpers-llm.js fetches the CHAT_MODELS configuration from the config table and parses it as JSON. If the CHAT_MODELS configuration doesn't exist, it falls back to a default array of chat models.
create or replace function match_memories (
query_embedding public.vector(1536),
match_threshold float,
match_count int
)
returns table (
id bigint,
key text,
value text,
user_id text,
related_message_id bigint,
similarity float
)
language sql stable
as $$
select
memories.id,
memories.key,
memories.value,
memories.user_id,
memories.related_message_id,
1 - (memories.embedding <=> query_embedding) as similarity
from memories
where 1 - (memories.embedding <=> query_embedding) > match_threshold
order by similarity desc
limit match_count;
$$;- DISCORD_PUBLIC_KEY: Visit the Discord Developer Portal, create a new application, and find your Public Key under the 'General Information' tab.
- DISCORD_APP_ID: This is the 'Client ID' found in the same section as above in the Discord Developer Portal.
- DISCORD_BOT_TOKEN: In the Discord Developer Portal, navigate to 'Bot', create a new bot, and use the token provided here.
- DISCORD_VOICE_CHANNEL_ID and DISCORD_GUILD_ID: These IDs can be obtained by enabling 'Developer Mode' in Discord settings, right-clicking on the channel/guild, and selecting 'Copy ID'.
- OPENAI_API_KEY: Sign up at OpenAI, and after verification, you can access your API key in your account settings.
- OPENAI_API_ORGANIZATION: This is your organization ID on OpenAI, available in your account settings on the OpenAI dashboard.
- GITHUB_USER: The bot's github username.
- GITHUB_TOKEN, GITHUB_APP_ID, GITHUB_INSTALLATION_ID, GITHUB_PERSONAL_ACCESS_TOKEN: Generate these tokens by going to your GitHub settings, selecting 'Developer settings' → 'Personal access tokens', and creating a new token with the appropriate scopes.
- GITHUB_PERSONAL_ACCESS_TOKEN: (Duplicate entry, see above)
- WOLFRAM_APP_ID: Sign up at Wolfram Alpha Developer Portal, create an application, and use the AppID provided.
- ELEVENLABS_VOICE_ID and ELEVEN_LABS_API_KEY: Register at ElevenLabs, create an application, and find your API Key and Voice ID under your application settings.
- GOOGLE_CLIENT_ID and GOOGLE_SECRET: Go to the Google Cloud Console, create a new project, and navigate to 'APIs & Services' > 'Credentials'. Here, you can create your OAuth 2.0 credentials. You will need to create a service account and download the JSON file containing your credentials. You will also need to enable the Google Drive, Calendar, and Docs APIs for your project.
- MASTODON_API_URL and MASTODON_ACCESS_TOKEN: Register your application with your Mastodon instance to receive the API URL, then follow the authentication process to obtain your access token.