Solana Agent

Why Solana Agent

Batteries Included

• Based on CyberChipped - a Python OpenAI Assistant Framework
• Provides conversational memory, parallel function calling, smart automatic tool choice, and message history using MongoDB
• Utilizes FastAPI and Next.js - the most popular and supported web frameworks
• Quickly add custom functions to your AI agent in a few lines of code
• Solana Agent Actions:
  • Transfer tokens from the Agent wallet to other wallets - via Helius
  • Swap tokens inside the Agent wallet - via Jupiter and Helius
• Social Integrations:
  • X - via the X API (Basic Plan)

Better than Eliza

• Solana Agent requires no-code changes, only adding simple environment variables, and MongoDB/Redis to work out of the box for a complete AI Agent with real-time chat
• Solana Agent's conversational history is superior to RAG for user interactivity, tool usage, and agent memory/recall/context
• Solana Agent's parallel tool calling and automatic AI tool choice is superior than using any LLM completion API with tool usage from any provider
• Solana Agent is an opinionated framework with one way to do things to keep things simple
• Solana Agent is written in Python the most popular language on GitHub and in the AI field
• Solana Agent replies on X in real-time and only replies when it should

Local Dev

Run locally on Mac or Linux

• Clone this repo - git clone https://github.com/truemagic-coder/solana-agent
• Ensure the latest LTS Node is installed with yarn
• Ensure Python 3.12.7 is installed with poetry
• Ensure Docker and Docker Compose are installed
• docker-compose up -d
• Rename .env.sample to .env in site and agent
• Get and set the PRIVATE_KEY (in base58 string format) in the site folder for .env file - if you want Solana Actions like sending tokens and swapping with funds from this wallet
• Get and set the OPENAI_API_KEY var in the agent folder for .env file - OpenAI API Keys
• Get and set the HELIUS_API_KEY and HELIUS_RPC_URL in both folders .env files - Helius
• If you want XBot - then setup a X developer account and setup the keys in agent folder for .env file - and uncomment the code in main.py
• Set all the secrets to match between the .env files and make them uuidv4s or other strong keys
• Open two terminal windows
• Terminal 1: cd site && yarn install && yarn dev
• Terminal 2: cd agent && poetry install && bash ./dev.sh
• Open your browser to http://localhost:3000

Deploy

Deploy to Heroku or Dokku

• Provision a MongoDB and Redis database
• Get one domains with two sub-domains - one for the site and one for the agent
• Add the proper env vars on Heroku or Dokku to your apps
• Add your proper remotes in each folder locally (each folder site and agent should be their own repos - git init)
• For each folder (site and agent) git commit and git push to the main branch
• Make sure to ps:scale worker and scheduler to 1 (web should already be 1)

Advanced Topics

Agent Functions

• Use expressive snake-case names for the functions with expressive param names (if required)
• Functions only take str params and must output a str (string)
• Functions must be fully sync - you cannot use async libraries or methods - example: using requests not httpx (sync vs async)
• Don't make tool outputs (strings) too large as when calling multiple calls in parallel has a size limit (combined)
• Keep in mind the 128k model token input limit when processing data especially from APIs
• OpenAI gpt-4o is recommended over gpt-4o-mini simply for AI IQ but if cost is a concern then mini may work for your agent usage

Production Apps

• WB AI Agent

Contributing

Contributions to Solana Agent are welcome! Please feel free to submit a Pull Request.

License

This project is licensed under the MIT License - see the LICENSE file for details.