feat: Add Daytona Deployment Support

[hyoungjun-noh]

Summary

This PR adds support for running SWE-ReX agents in Daytona sandbox environments. Daytona is a sandbox environment platform that provides secure, isolated execution environments for AI agents.

Changes

New Components

• DaytonaRuntime (runtime/daytona.py): SDK-based runtime that communicates with the swerex server running inside a Daytona sandbox. Uses the Daytona SDK's command execution capabilities instead of HTTP requests to bypass OAuth authentication issues with Preview URLs.

• DaytonaDeployment (deployment/daytona.py): Deployment class for creating and managing Daytona sandboxes. Supports both self-hosted Daytona servers and public Daytona cloud.

Key Features

• SDK-based communication: Uses Daytona SDK's execute_session_command to run commands inside the sandbox, avoiding OAuth authentication issues with Preview URLs
• Self-hosted and cloud support: Configure with api_url for self-hosted or target for Daytona cloud
• Health check polling: Monitors swerex server startup inside the sandbox using Python's urllib (no curl dependency)
• Base64 encoding: Uses base64 encoding for command transfer to avoid shell escaping issues
• Session management: Creates separate sessions for server process, health checks, and runtime communication

Configuration

from swerex.deployment.config import DaytonaDeploymentConfig
from swerex.deployment.daytona import DaytonaDeployment

# Self-hosted Daytona server
config = DaytonaDeploymentConfig(
    image="python:3.12-slim",
    api_url="https://your-daytona-server.com",
    api_key="your-api-key",
    port=8000,
)

# Or Daytona cloud
config = DaytonaDeploymentConfig(
    image="python:3.12-slim",
    target="us",
    api_key="your-api-key",
)

deployment = DaytonaDeployment.from_config(config)
await deployment.start()

Installation

pip install 'swe-rex[daytona]'

Implementation Details

Communication Flow

1. DaytonaDeployment creates a Daytona sandbox from the specified image
2. Starts the swerex server inside the sandbox using SDK command execution
3. Polls for server health using Python urllib (no curl dependency)
4. Creates DaytonaRuntime with a separate session for runtime commands
5. DaytonaRuntime uses SDK command execution with base64-encoded Python scripts to call swerex API endpoints

Why SDK-based Communication?

Direct HTTP access to the swerex server via Preview URLs encounters OAuth authentication issues. By using the SDK's command execution to run Python scripts inside the sandbox, we can communicate with the swerex server on localhost without going through the OAuth layer.

Testing

Tested with:

• Self-hosted Daytona server and public Daytona cloud service
• Mini-SWE-Agent running terminal-bench tasks
• Multiple concurrent sandbox creation and execution

Files Changed

• src/swerex/runtime/daytona.py - New DaytonaRuntime class
• src/swerex/deployment/daytona.py - New DaytonaDeployment class
• src/swerex/deployment/config.py - Added DaytonaDeploymentConfig
• pyproject.toml - Added daytona-sdk dependency and [daytona] extra
• README.md - Updated installation instructions
• CHANGELOG.md - Added Unreleased section with Daytona support

Related

• Daytona Documentation
• Daytona Python SDK