Ducky

Read the Intro blog!

Ducky is an AI agent operating the twitter account Duck Unfilitered You can see his stream of thought here Ducky Website

Warning

This code is not Stable, in the midst of a refactoring and heavy developement, do not rely on it for mission critical stuff, yet. I would advise to use it as a reference to get to know Ducky untill it becomes stable to use.

This code runs the Unfiltered Duck, Ducky on twitter.

• Character Prompt: Character
• Agents - Context Injecting prompts
• Wallets - AI controlled Turnkey TEE wallet, cosigned on a gnosis Safe multisig Code
• UI - Website (Stream of thought backend logs, community dashboard, chat v2)
• Sentiment Analysis - Tracks sentiment in telegram and twitter replies
• Conversations - (in progress), personalized chat with Ducky

Getting Started

1. We use Railway.app
2. Create a postgres service
3. Frontend Service: Root Directory: /frontend, custom start command bun start
4. Discord Service: ./entrypoint.sh discord
5. Telegram Service: ./entrypoint.sh telegram
6. Cleo Tweeter: (From conversations with Cleo, see blog Root Dir: twitter-server and Custom Start Command: bun generateCleoConvos
7. Degen Tweeter: Trained on his previous tweets and responses. Root Dir: twitter-server and Custom Start Command: bun generateTweet
8. Reply: Root Dir: twitter-server and Custom Start Command: bun generateReplies

Env

ANTHROPIC_API_KEY
DATABASE_URL
TODO:(some others not documented yet)

Structure

This repo is part python, javascript and rust. Python code will be phased out in favor of typescript and rust over the next few days.

├── conversations - rust server for conversations (not in production yet)
├── db - Database (Depreciating in favor of more typesafe in Drizzle)
├── memory - database
├── frontend - next.js frontend app for ducky
├── ducky - main typescript brain
├── utils/sentiment_analysis/message_fetcher - tg messages backfill
├── wallet - turnkey and gnosis

Drizzle Studio

./entrypoint.sh studio

Hosting

Runpod Llama3.1:70b

• Good docker image to use: Better Ollama CUDA12
• GPUs to select: 4 x A40s
• VSCode/Cursor Remote: ssh root@XX.XX.XX.XXX -p 22XXX -i ~/.ssh/<RUNPOD_SSH>
• Install ollama: (curl -fsSL https://ollama.com/install.sh | sh && OLLAMA_HOST=0.0.0.0 ollama serve > ollama.log 2>&1) &
• KeepAlive: ollama run llama3.1:70b --keepalive 1000m

DONT FORGET OLLAMA_HOST=0.0.0.0 in above command

Twitter Rate Limit (for reference)

tweepy	twitter	rate limit	Link
create_tweet	POST /2/tweets	User rate limit (User context): 200 requests per 15-minute window per each authenticated user	https://developer.x.com/en/docs/x-api/tweets/manage-tweets/api-reference/post-tweets
search_recent_tweets	GET /2/tweets/search/recent	App rate limit (Application-only): 450 requests per 15-minute window shared among all users of your app	User rate limit (User context): 180 requests per 15-minute window per each authenticated user

Acknowledgement

• agent-twitter-clientsaved us from ERROR 429 - Elon hates you, errors
• Luna Virtuals - awesome brain and prompting