Guides
The guides cover a wide range of topics, from basic usage to advanced topics, and showcase various types of integrations between Daytona and other tools.
Filters
Build a Coding Agent Using Letta Code and Daytona
Step-by-step guide to running Letta Code, a stateful coding agent with persistent memory, inside Daytona sandboxes.
- Letta Code
Fix Bugs Automatically With AG2 and Daytona
Build a multi-agent bug fixer that analyzes broken code, proposes fixes, and verifies them in isolated Daytona sandboxes.
- AG2
Build an Autonomous Bug-Fix Agent with Flue and Daytona
Use Flue's agent harness and a Daytona sandbox to reproduce a GitHub issue with a failing test, fix it, and open a real pull request, end to end.
- Claude
Build a Generative-UI Coding Agent with CopilotKit and Daytona
Wire a CopilotKit Built-in Agent to a Daytona sandbox with full shell and filesystem access, where every tool call streams into the chat as generative UI and any hosted process embeds as a live iframe.
- CopilotKit
Serve LLMs on GPU Sandboxes with SGLang
Serve OpenAI's gpt-oss reasoning model on a Daytona GPU sandbox with SGLang, and put it to work classifying documents with guaranteed-structure output.
- Model Serving
Serve LLMs on GPU Sandboxes with vLLM
Spin up an OpenAI-compatible vLLM endpoint on a Daytona GPU sandbox and query it from anywhere.
- Model Serving
Run an RL Coding Environment at Scale with HUD and Daytona
Define a coding environment with HUD, scale it to 256 parallel rollouts on Daytona sandboxes, and train a model on the graded rollouts.
- HUD
Multi-Turn RL Training with OpenEnv and Daytona
Use parallel Daytona sandboxes with OpenEnv's FinQA environment to train Qwen3-14B on financial question answering via GRPO.
- TRL
- OpenENV
Train LLMs with Reinforcement Learning Using TRL and Daytona
Use parallel Daytona sandboxes to train code-generating LLMs with TRL's GRPO trainer.
- HUD
Scaling Tool Execution in RL Training with veRL and Daytona
Run veRL's ReTool recipe on Daytona sandboxes, configure the Daytona backend, and benchmark tool-calling throughput.
- HUD