JSPM

  • ESM via JSPM
  • ES Module Entrypoint
  • Export Map
  • Keywords
  • License
  • Repository URL
  • TypeScript Types
  • README
  • Created
  • Published
  • Downloads 481
  • Score
    100M100P100Q109747F
  • License MIT

Scaffold and run a Balchemy autonomous AI trading agent in 5 minutes

Package Exports

  • balchemy
  • balchemy/dist/index.js

This package does not declare an exports field, so the exports above have been automatically detected and optimized by JSPM instead. If any package subpath is missing, it is recommended to post an issue to the original package (balchemy) to support the "exports" field. If that is not possible, create a JSPM override to customize the exports field for this package.

Readme

Balchemy

Autonomous AI Trading Agent Platform
Deploy a dual-LLM trading agent on Solana and EVM chains in 5 minutes.

npm version npm downloads node version license platform


What Is Balchemy?

Balchemy is an autonomous AI trading platform that connects your LLM to on-chain markets through the Model Context Protocol (MCP). You provide the strategy in natural language; your LLM decides when and what to trade; Balchemy handles wallets, execution, risk checks, and 100+ trading tools.

Architecture — Dual-LLM System:

  • External LLM (your choice: Claude, GPT, Gemini, Grok, OpenRouter) — the brain that makes all decisions
  • Inner LLM (GPT-5.4-mini, server-side) — infrastructure servant that fetches data, formats responses, and serves the external LLM

Your LLM calls Balchemy tools via MCP — it picks the right tool from natural language descriptions, not hardcoded tool names.

You (strategy) → External LLM (decisions) → Balchemy MCP (execution) → Solana / Base / Ethereum
                                  ↓
                        Inner LLM (data fetching, formatting)

Quick Start

npx balchemy

The interactive wizard walks you through:

  1. Pick your LLM — Anthropic, OpenAI, Gemini, Grok, or OpenRouter
  2. Set up wallets — auto-provisioned Solana and EVM wallets
  3. Define your strategy — natural language rules ("max 5% position size, only trade on Solana")
  4. Start the 24/7 agent loop — your LLM monitors markets and trades autonomously
npx balchemy          # Setup wizard or resume cached agent
npx balchemy init      # Force new setup wizard
npx balchemy start     # Start from agent.config.yaml
npx balchemy docker    # Generate Docker files for deployment

What You Can Do

Trade

  • Buy/Sell tokensbuy 0.1 SOL worth of BONK on Solana
  • Limit orders — set price targets, get filled automatically
  • DCA (Dollar-Cost Averaging) — schedule recurring buys
  • Trailing stops — lock profits as price moves

Research

  • Token analysis — market cap, volume, holder distribution, risk scores
  • Smart money tracking — follow profitable wallets
  • Bundle detection — spot insider token launches
  • Price queries — real-time Solana and EVM token prices

Manage

  • Portfolio view — see balances, P&L, positions across chains
  • Behavior rules — define trading constraints in natural language
  • Subscriptions — set up alerts for market events
  • 24/7 autonomous loop — your LLM monitors and acts even when you're away

Risk Management (Built-In)

Every trade passes through multi-layer safety checks before execution:

  • RugCheck integration — community risk scores
  • Honeypot detection — simulated sell before buying
  • Contract verification — liquidity locks, ownership renouncing
  • Behavior rules enforcement — your constraints are always respected

100+ MCP Tools

Your LLM accesses Balchemy through these primary agent-facing tools:

Tool What It Does
trade_command Execute buy/sell/swap on Solana or EVM
ask_bot Natural language market queries (inner LLM-powered)
agent_research Deep token research with technical analysis
agent_portfolio Portfolio, positions, P&L overview
configure_behavior_rules Set trading constraints in natural language
get_behavior_rules Read current active rules
create_subscription Set up market event alerts
setup_agent Wallet provisioning and onboarding

Plus 90+ internal tools for research, risk scoring, and execution — your LLM picks the right one from descriptions.

LLM Providers

Provider Environment Variable Notes
Anthropic ANTHROPIC_API_KEY Haiku 4.5, Sonnet 4.6, Opus 4.6+
OpenAI OPENAI_API_KEY GPT-4o, GPT-5.4, etc.
Google Gemini GEMINI_API_KEY Gemini 3.1 Pro, Flash
Google Vertex AI GOOGLE_APPLICATION_CREDENTIALS Gemini 3.1 Pro via Vertex AI
xAI Grok GROK_API_KEY Grok-4-1-fast for research
OpenRouter OPENROUTER_API_KEY Access to 100+ models

Configuration

After setup, agent.config.yaml and .env are generated in the current directory:

# agent.config.yaml
llm:
  provider: anthropic
  model: claude-sonnet-4-20250514

agent:
  name: my-trading-agent
  strategy: Focus on Solana memecoins with high volume and positive sentiment

wallets:
  solana:
    chain: solana
  evm:
    chainId: 8453  # Base

Edit these files to change settings without re-running the wizard.

24/7 Autonomous Mode

The SDK includes an AgentLoop that keeps your LLM connected to Balchemy permanently:

import { BalchemyAgentSdk, AgentLoop } from "@balchemyai/agent-sdk";

const sdk = new BalchemyAgentSdk({ apiBaseUrl: "https://api.balchemy.ai/api" });
const loop = new AgentLoop(sdk, { apiKey: "your-mcp-api-key" });

// Starts SSE event stream + polling cycle
// Your LLM receives market events and decides whether to act
await loop.start();

Events flow: Market event → SSE push → Your LLM evaluates → trade_command if relevant → Execution → Confirmation push

Hub Integration

When you bind your EVM wallet during setup, your agent appears in your Balchemy Hub dashboard:

  • Monitor — live portfolio, trade history, event log
  • API Keys — create, rotate, and revoke MCP keys with scoped permissions
  • Scope Managementread for data-only, trade for full execution
  • Wallet Management — link/unlink Solana and EVM wallets

Security

  • Encrypted credentials — AES-256-GCM at rest with PBKDF2 key derivation
  • Scoped API keysread and trade permissions, instant revocation
  • No seed phrases stored — all keys encrypted, never logged
  • Behavior rules — hard constraints that even your LLM cannot override
  • Pre-trade safety checks — RugCheck, honeypot detection, contract verification

Requirements

  • Node.js 18+
  • An LLM API key from any supported provider

License

MIT