Condor’s Two-Layer Architecture for Autonomous Trading Agents
Summary
Condor is presented as an open-source framework for building autonomous trading agents. Its architecture separates an agent layer, where large language models observe conditions and make decisions through an Observe, Orient, Decide, Act loop, from an execution layer that manages positions, trading operations, and risk enforcement through the Hummingbot API. This separation aims to keep model-driven decisions distinct from the more deterministic mechanisms that carry out and track trades.
The documentation describes virtual positions for spot, perpetual futures, and liquidity-provider exposure, along with trading bots, deterministic routines, and access through messaging, a dashboard, or a command-line interface. It also outlines setup and integrations, but provides no trading strategy, live performance evidence, or evaluation of agent reliability. The framework’s capabilities therefore describe infrastructure rather than proof that autonomous agents can trade profitably or safely. The page directs readers elsewhere for detailed installation instructions and technical specifications.
Key ideas
- Condor separates language-model decision-making from deterministic trade execution and risk enforcement.
- Agents use an Observe, Orient, Decide, Act loop to make decisions over successive ticks.
- The framework supports virtual spot, perpetual-futures, and liquidity-provider positions.
- Bots, routines, and several user interfaces provide ways to run and monitor trading operations.
- The documentation describes infrastructure and integrations but provides no evidence of trading performance or agent reliability.
Tags
Full text
# Condor
# Condor
**Condor** is an open source harness for building and running autonomous **Trading Agents**. It connects LLM-powered decision-making to deterministic trade execution via the [Hummingbot API](../hummingbot-api/index.md), enabling traders to deploy AI agents that can observe markets, reason about strategy, and execute trades across 50+ exchanges and blockchains.
!!! note "Full Documentation"
The complete Condor documentation has moved to a dedicated site. Visit **[condor.hummingbot.org](https://condor.hummingbot.org)** for full installation guides, Trading Agents Standard specification, and API reference.
--8<-- "docs/includes/condor-feedback.md"
<iframe style="width:100%; min-height:400px;" src="https://www.youtube.com/embed/O93R_ddB-8o" frameborder="0" allow="accelerometer; autoplay; encrypted-media; gyroscope; picture-in-picture" allowfullscreen></iframe>
## What is Condor?
Condor implements a two-layer architecture that separates probabilistic reasoning from deterministic execution:
- **Agentic Layer**: LLM-powered reasoning using the OODA loop (Observe, Orient, Decide, Act)
- **Execution Layer**: Deterministic infrastructure for positions, executors, and risk enforcement via Hummingbot API
Where Hummingbot lets one person do the work of a team, Condor lets one person manage a *swarm* of autonomous agents—each observing markets, adapting to conditions, and executing strategies independently.
## Key Features
| Feature | Description |
|---------|-------------|
| **Trading Agents** | AI-powered agents that make decisions each tick using LLMs (Claude, Gemini, Codex) |
| **Executors** | Self-contained trading operations with standardized P&L tracking |
| **Positions** | Virtual portfolio tracking for spot, perp, and LP positions |
| **Bots** | Docker containers for long-running market making and grid trading |
| **Routines** | Deterministic workflows for indicators, webhooks, and alerts |
| **Multi-Interface** | Telegram bot, web dashboard, and CLI access with session continuity |
## Quick Start
Install Condor with a single command:
```bash
curl -fsSL https://raw.githubusercontent.com/hummingbot/deploy/main/setup.sh | bash
```
This script clones the repository, installs dependencies, and asks how you want
to drive Condor:
- **Telegram** (recommended) — it prompts for your Telegram Bot Token and User ID
- **Local** — no Telegram at all; the web dashboard runs on that machine at
`http://localhost:8088` with no login and a loopback-only bind
It then walks you through Tailscale, your AI model, and the Hummingbot API
connection. See [Condor Quickstart](../installation/condor.md) for the full
prompt-by-prompt walkthrough, and run `make doctor` afterwards to verify the
install.
## Resources
| Resource | Link |
|----------|------|
| Full Documentation | [condor.hummingbot.org](https://condor.hummingbot.org) |
| GitHub Repository | [github.com/hummingbot/condor](https://github.com/hummingbot/condor) |
| Give Feedback | [2-minute survey](https://forms.gle/7NpG3RtgfLrmpUNY8) |
| Discord Support | [#condor-feedback](https://discord.gg/hummingbot) |
| Hummingbot API | [Hummingbot API Docs](../hummingbot-api/index.md) |Shown in full with attribution under the source's licence. Licence: Apache-2.0
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.