Skip to main content

​
What is Condor?

Condor is an open source agent harness for deploying and managing fleets of AI trading agents. It connects LLM-powered decision-making to deterministic trade execution, enabling you to run autonomous agents that observe markets, reason about strategy, and execute trades.
Everything in Condor is free and open source. We’re supported by exchanges as partners, so you can use everything without paying.
New to Condor? Head to the Learning Path to find the setup and reading order that matches your experience level and use case—whether that’s letting AI trade for you, running an always-on bot, or deploying to production.

Help shape Condor

What’s working? What’s confusing? What do you wish Condor could do? Your answers go straight to the Hummingbot Foundation team and directly shape what we build next.

​
What can you do with it?

Scale your trading operation
  • Run dozens of agents simultaneously on shared accounts
  • Each agent tracks its own P&L independently
  • Experiment with different strategies in parallel
Trade anywhere
  • 50+ CEX connectors (Binance, Bybit, OKX, Hyperliquid, etc.)
  • DEX support via Gateway (Uniswap, Jupiter, Raydium)
  • Multi-leg strategies across exchanges and chains
Choose your execution style
  • Trading Agents: AI-powered, make decisions each tick
  • Bots: Long-running containers for market making, grid trading
  • Executors: Single trades with defined entry/exit

​
Who is it for?

Algorithmic traders who want to add AI reasoning to their strategies without rebuilding execution infrastructure. Quants and researchers who want to experiment with many agents, compare performance, and iterate quickly. Developers who want to build trading applications on top of a battle-tested execution layer. Crypto funds who need multi-agent orchestration with proper P&L isolation and audit trails.

​
What makes it different?

Two-server architecture separates LLM reasoning from trade execution. The Condor Server handles agent logic and state; the Hummingbot API handles data and execution. This means:
  • Execution continues even if the LLM is slow
  • Agents can share market data efficiently
  • You can swap LLM providers without touching execution code
Full observability. Every tick is captured: the prompt sent to the LLM, its reasoning, tool calls made, and results. You can replay any decision. Deterministic routines move expensive computations (indicators, data processing) out of the LLM into reusable Python code. This cuts token costs and makes behavior reproducible. Multi-agent by design. Agents use controller_id to isolate their positions and P&L on shared exchange accounts. Run 50 agents on one API key.

​
Who is behind this?

Condor is built by Hummingbot Foundation, a not-for-profit organization that maintains the open source Hummingbot trading framework used by thousands of traders worldwide. The Foundation is supported by exchange partners and community contributions. There’s no VC funding or token—just open source software.

​
How do I get started?

Learning Path

Find your route by experience level and use case

Motivation

Understand why Condor is architected this way

Installation

Install Condor and Hummingbot API

Getting Started

Build your first trading agent

GitHub

Source code and issues

Give Feedback

2-minute survey that shapes the roadmap

​
Reading these docs with an agent

These docs are published in a machine-readable form, so you can hand them to an LLM rather than paraphrasing them at it: Any page also serves as Markdown by appending .md to its URL — for example /getting-started/installing.md.
Condor’s own agents already read the parts of this that matter to them. This is for your assistant — pointing Claude, Codex or Cursor at /llms.txt is usually enough for it to answer setup questions against the current docs instead of its training data.