Enum ToolExecutionResult 

pub enum ToolExecutionResult {
    Success(Value),
    SuccessWithImages {
        result: Value,
        images: Vec<ToolResultImage>,
    },
    ToolError(String),
    InternalError(ToolInternalError),
    ConnectionRequired {
        provider: String,
    },
}

Result of a tool execution.

This enum distinguishes between different outcomes:

• Success: Tool executed successfully, result is returned to LLM
• SuccessWithImages: Successful execution with JSON result plus images
• ToolError: Tool-level error that should be shown to the LLM (e.g., “City not found”, “Invalid date format”)
• InternalError: System-level error that should NOT be exposed to the LLM (e.g., database connection failure, API key issues)

Security

Internal errors are logged but replaced with a generic message when returned to the LLM. This prevents leaking sensitive information like database errors, API keys, or internal system details.

Variants

Success(Value)

Successful execution with a JSON result

SuccessWithImages

Successful execution with a JSON result and images. Images are sent to the LLM as native image content blocks (not stringified JSON), enabling visual understanding.

Fields

result: Value

images: Vec<ToolResultImage>

ToolError(String)

Tool-level error that is safe to show to the LLM

Use this for expected error conditions that the LLM should know about, such as validation errors, resource not found, etc.

InternalError(ToolInternalError)

Internal/system error that should NOT be exposed to the LLM

Use this for unexpected errors like network failures, database errors, or other internal issues. The error details will be logged but replaced with a generic message when returned to the LLM.

ConnectionRequired

A user connection is required to execute this tool.

Instead of returning an error, this signals that the workflow should pause and ask the client to set up a connection for the given provider. The UI renders an inline connection dialog; once the user saves (or cancels), a tool result is submitted and execution resumes.

Fields

provider: String

Connection provider id (e.g. “daytona”, “brave_search”)

Implementations

impl ToolExecutionResult

pub fn success(value: impl Into<Value>) -> Self

Create a successful result

pub fn success_with_raw_output( value: impl Into<Value>, raw_output: String, ) -> Self

Create a successful result with pre-truncation raw output for VFS persistence. The raw output is transferred to ToolResult.raw_output during into_tool_result().

pub fn success_with_images( value: impl Into<Value>, images: Vec<ToolResultImage>, ) -> Self

Create a successful result with images

pub fn tool_error(message: impl Into<String>) -> Self

Create a tool-level error (safe to show to LLM)

pub fn internal_error(error: impl Error + Send + Sync + 'static) -> Self

Create an internal error (will be hidden from LLM)

pub fn internal_error_msg(message: impl Into<String>) -> Self

Create an internal error from a string message

pub fn connection_required(provider: impl Into<String>) -> Self

Signal that a user connection is required before this tool can execute.

pub fn is_success(&self) -> bool

Check if this is a successful result

pub fn is_error(&self) -> bool

Check if this is an error (either tool error or internal error)

pub fn is_connection_required(&self) -> bool

Check if this requires a user connection setup

pub fn into_tool_result(self, tool_call_id: &str, tool_name: &str) -> ToolResult

Convert to a ToolResult for the agent loop

Both tool errors and internal errors are packaged as {"error": "..."} in the result field. This provides a consistent contract where the result field always contains the payload, and the agent loop continues the same way for all outcomes.

Internal errors are logged but replaced with a generic message when returned.

Trait Implementations

impl Debug for ToolExecutionResult

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.