> ## Documentation Index
> Fetch the complete documentation index at: https://docs.agenticenv.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Code Execution

> Let the LLM write and run code in an isolated sandbox via a custom execute_code tool

Shows how to give an agent a code execution capability using a custom `execute_code` tool. The LLM writes a short script, the sandbox runs it, and the output is returned for the LLM to explain. The sandbox is pluggable — swap the runtime without changing the tool.

Source: [`examples/agent_with_code_execution/`](https://github.com/agenticenv/agent-sdk-go/tree/main/examples/agent_with_code_execution)

## What it demonstrates

* `execute_code` custom tool — `interfaces.Tool` with `language` and `code` parameters
* `SandboxRuntime` interface — pluggable sandbox (local os/exec, Docker, cloud APIs)
* LLM writes code, runs it, reads output, replies — never makes up results
* Sandbox isolation — Docker runner uses `--network none` and memory limits

## Run

From `examples/`:

Local sandbox — needs Python or Node installed on host:

```bash theme={null}
go run ./agent_with_code_execution "Write a Python script that prints the first 10 Fibonacci numbers"
```

Docker sandbox — isolated, no host interpreter needed:

```bash theme={null}
SANDBOX_ENV=docker go run ./agent_with_code_execution "Write a Python script that prints the first 10 Fibonacci numbers"
```

Requires `LLM_APIKEY` in `examples/.env`.

## Key code

```go theme={null}
// newSandbox picks the execution environment from SANDBOX_ENV (default: local os/exec).
// Swap newLocalRunner for newDockerRunner, a Lambda runner, Modal, E2B, etc.
sandbox, err := newSandbox()

a, err := agent.NewAgent(
    agent.WithLLMClient(llmClient),
    agent.WithTools(NewCodeTool(sandbox)),
    agent.WithAgentMode(agent.AgentModeAutonomous),
    agent.WithToolExecutionConfig(agent.ExecutionConfig{
        Timeout:     2 * time.Minute,
        MaxAttempts: 1,
    }),
)

result, err := a.Run(context.Background(), prompt, nil)
```

`Execute` blocks until `SandboxRuntime.Execute` returns the program output and exit code.

## Expected output

```
user: Write a Python script that prints the first 10 Fibonacci numbers

A: Here are the first 10 Fibonacci numbers: 0, 1, 1, 2, 3, 5, 8, 13, 21, 34.
```

## Learn more

<CardGroup cols={2}>
  <Card title="Code Execution in Agents" icon="terminal" href="/advanced/code-execution" horizontal>
    Sandbox options, isolation, timeout config, cloud runtimes
  </Card>

  <Card title="Execution config" icon="sliders" href="/features/execution-config" horizontal>
    Tool execute timeout and max attempts
  </Card>
</CardGroup>