Runnable, self-contained examples showing how to build AI agents, data analysts, model-serving endpoints, and reinforcement-learning workflows on Daytona sandboxes.
Each guide lives in its own folder with its own README, dependencies, and .env.example. Most guides also have a step-by-step walkthrough in the Daytona docs.
- A Daytona account and API key — create one in the Daytona Dashboard.
- The runtime the guide uses — Python or Node.js. Minimum versions vary by guide (baseline Python 3.10+ / Node.js 18+, though some need newer, e.g. Python 3.12 or Node.js 22), so check the guide's own README.
- Any provider API keys a given guide needs (OpenAI, Anthropic, Gemini, etc.) — see that guide's README.
git clone https://github.com/daytona/guides.git
cd guides/<path-to-guide> # e.g. typescript/openai/codex-sdk
cp .env.example .env # then fill in your keysThen follow the guide's own README to install dependencies and run it.
| Guide | Description |
|---|---|
| AG2 Bug Fixer | Two AG2 agents that iteratively fix broken code, running each attempt in a sandbox. |
| Data Analyst (LiteLLM) | Natural-language data analysis with LiteLLM, executing Python in a sandbox. |
| Data Analyst (OpenAI) | Natural-language data analysis with the OpenAI API, executing Python in a sandbox. |
| Claude Managed Agents | Self-host Claude Managed Agents inside Daytona sandboxes. |
| Windows Computer-Use Evals | Windows GUI computer-use eval suite driven by Claude Code / Codex via MCP. |
| DSPy RLMs | Run DSPy's Recursive Language Model (REPL-in-a-loop) module on Daytona. |
| Google ADK Code Generator | Google ADK agent that generates and verifies code with the Daytona plugin. |
| LangChain Data Analysis | LangChain agent doing secure data analysis via the Daytona data-analysis tool. |
| LangGraph Plan-and-Execute | Plan-and-execute ETL + analytical-SQL agent wired as a six-node state machine. |
| Model Serving (SGLang) | Serve gpt-oss-20b with SGLang on a GPU sandbox behind a preview URL. |
| Model Serving (vLLM) | Serve an open-weights model with vLLM on a GPU sandbox behind a preview URL. |
| Recursive Language Models | Recursive-LM agents that spawn sub-agents, each in its own sandbox. |
| OpenEnv (FinQA) | Evaluate and train models on FinQA with OpenEnv and Daytona sandboxes. |
| TRL RL Rollouts | Run TRL RL rollouts, executing generated code in parallel sandboxes. |
| veRL ReTool Benchmark | Benchmark script for veRL's ReTool sandbox execution backend. |
| Guide | Description |
|---|---|
| Inngest AgentKit Coding Agent | Autonomous coding agent built with Inngest AgentKit. |
| Data Analyst (OpenAI) | Natural-language data analysis with the OpenAI API, executing code in a sandbox. |
| Amp Code Coding Agent | Coding agent powered by the Amp Code CLI. |
| Claude Two-Agent System | Two-agent Claude system coordinating work across sandboxes. |
| Claude Coding Agent | Single Claude Code agent you drive from the CLI. |
| Devin CLI Coding Agent | Coding agent powered by Cognition's Devin CLI. |
| CopilotKit Generative-UI Agent | CopilotKit agent that streams every tool call as generative UI. |
| Flue Bug-Fix Agent | Autonomous GitHub-issue bug-fix agent built with Flue. |
| Gemini CLI Coding Agent | Headless coding agent powered by the Gemini CLI. |
| Kiro CLI Coding Agent | Coding agent powered by AWS Kiro's CLI. |
| Letta Code Agent | Letta Code agent driven from the CLI. |
| Mastra Code Execution Agent | Mastra agent that plans, writes, executes, and iterates on code. |
| OpenAI Codex Agent | OpenAI Codex agent driven from the CLI. |
| OpenClaw Assistant | OpenClaw general-purpose AI assistant with a web Control UI. |
| OpenCode Server | OpenCode coding-agent server driven from the CLI. |
| OpenCode Web | OpenCode agent with a web interface. |
| Vercel AI SDK Benchmark Agent | Vercel AI SDK agent that benchmarks Python vs TypeScript and plots the results. |
Each guide is self-contained. Keep changes scoped to a single guide's folder, include a clear README, and provide a .env.example for any required keys.
Apache-2.0 © Daytona Platforms Inc.