Module code_exec 

Code-execution subsystem (feature = "code-exec").

Lets agents emit ExecutableCode parts and have the runtime run the code in a sandbox of varying degrees of safety.

• local::LocalCodeExecutor — spawns a child interpreter via tokio::process with a timeout. Local subprocess isolation only — not a security boundary.
• docker::ContainerCodeExecutor (feature = "code-exec-docker") — spins up an ephemeral Docker container per call.

The runner extracts ExecutableCode parts from model responses, invokes the agent’s executor, and feeds back CodeExecutionResult parts. Attach an executor via LlmAgent::builder().code_executor(...).

Modules

docker

ContainerCodeExecutor — runs code in a fresh ephemeral Docker container per call.

local

LocalCodeExecutor — runs code in a local subprocess via tokio::process.

Structs

CodeExecutionInput

Inputs to one execution.

CodeExecutionResult

Output of one execution.

ExecFile

One file passed to or returned from an executor.

Traits

CodeExecutor

Pluggable executor for ExecutableCode parts.