Remote Agents with Daytona

By default, Kiwi Code runs commands on your local machine through the CLI runtime. With Daytona, you can run code in isolated cloud sandboxes instead — giving you security, configurable compute, and the ability to run untrusted code safely.

When to Use Remote Sandboxes

Use Case	Why Remote
Untrusted or experimental code	Sandboxes are fully isolated — nothing touches your machine
Heavy compute	Need 8 cores and 16 GB RAM? Configure the sandbox
Long-running jobs	Let it run in the cloud without blocking your terminal
Browser automation & UI testing	VNC desktop support for graphical environments
Reproducible environments	Clean sandbox every time, no local dependency conflicts
CI/CD-style workflows	Run, collect artifacts, tear down

How to Use Daytona

From the Web Dashboard

In the web dashboard at kiwicode.ai, prompt the AI to use a remote sandbox:

Run this Python script in a Daytona sandbox with 4 cores and 8 GB RAM:
[paste your script or describe the task]

Create a sandbox and run the test suite for this project.
Collect the coverage report as an artifact.

Open a browser in a Daytona sandbox and screenshot the homepage of example.com

The AI handles sandbox creation, code execution, and artifact collection automatically.

Checking Job Status

From the TUI or CLI, you can check the status of remote jobs:

# List your Daytona jobs
kiwicli runs list --status processing

# Get details on a specific run
kiwicli runs get <run_id>

What Happens Behind the Scenes

  You (prompt)
    │
    ▼
  Kiwi AI (cloud)
    │
    ├─ Creates Daytona sandbox
    ├─ Executes code inside sandbox
    ├─ Collects output files (artifacts)
    └─ Returns results to you

The AI provisions an isolated cloud sandbox with the requested resources
Code runs inside the sandbox with full terminal access
Output files are collected and made available to you
The sandbox auto-stops when done (or can be kept alive for follow-up work)

Compute Options

Request different levels of compute based on your task:

Profile	CPU	Memory	Good For
Default	1 core	2 GB	Simple scripts, linting
Small	2 cores	4 GB	Unit tests, small builds
Medium	4 cores	8 GB	Integration tests, compilation
Large	8 cores	16 GB	ML training, large builds

Ask for what you need in your prompt:

Run this training script in a large sandbox (8 cores, 16 GB)

VNC Desktop

For tasks that need a graphical environment, Daytona sandboxes can provide a VNC desktop accessible from your browser:

Open a Daytona sandbox with VNC and automate the signup flow on staging.example.com

The AI gets a full Linux desktop with a browser, enabling visual automation and UI testing.

Artifacts

Files created during remote execution are automatically collected and made available to you. Common artifact types:

Test coverage reports
Build outputs
Screenshots and recordings
Generated files and exports
Log files

Local vs Remote — Which to Choose

	Local Runtime	Daytona Sandbox
Speed	Instant startup	A few seconds to provision
Isolation	Runs on your machine	Fully isolated cloud VM
Compute	Your hardware	Configurable CPU/memory
Best for	Interactive coding, file editing	Heavy compute, untrusted code, CI/CD
Files	Direct access to your filesystem	Artifacts uploaded after execution

Daytona is available from the web dashboard

Remote sandbox execution is orchestrated by the backend. Use the web dashboard to trigger Daytona runs. The TUI connects your local runtime for interactive coding.

Remote Agents with Daytona

On this page