Enum AgentEvent 

pub enum AgentEvent {
    TextDelta {
        content: String,
    },
    ReasoningDelta {
        content: String,
    },
    ToolCallStart {
        tool_call: ToolCall,
        display_name: String,
        args_summary: Option<String>,
    },
    ToolCallComplete {
        tool_name: String,
        result: ToolResult,
        wait_ms: u64,
        execute_duration_ms: u64,
        display_name: String,
        duration_display: String,
    },
    IterationComplete {
        iteration: usize,
        tool_calls: usize,
    },
    FinalAnswer {
        content: String,
    },
    PromptProgress {
        processed: u32,
        total: u32,
        cached: u32,
        time_ms: u64,
    },
    SystemWarning {
        message: String,
        suggested_action: Option<String>,
    },
    Error {
        message: String,
    },
}

An observable event emitted by the agentic loop.

These events are the unit of SSE emission: every state change in the loop produces exactly one variant. Axum SSE handlers serialise these to data: <json>\n\n frames; CLI consumers may log or render them directly.

Serde tag

#[serde(tag = "type", rename_all = "snake_case")] produces e.g. {"type":"tool_call_start","tool_call":{...}}.

Variants

TextDelta

An incremental text fragment from the model’s response.

Fields

content: String

The new text fragment (append to the current buffer).

ReasoningDelta

An incremental reasoning/thinking fragment from the model (CoT tokens).

Emitted by reasoning-capable models (e.g. DeepSeek R1, QwQ) that expose their chain-of-thought via a separate reasoning_content SSE field. These fragments are forwarded to the UI as they arrive but are not included in the conversation history sent back to the model.

Fields

content: String

The new reasoning fragment (append to the current reasoning buffer).

ToolCallStart

The model has requested execution of a tool.

Fields

tool_call: ToolCall

The tool call that is about to be dispatched.

display_name: String

Human-readable tool name (prefix stripped, title-cased).

args_summary: Option<String>

One-line argument summary (e.g. file path, search pattern).

ToolCallComplete

A tool execution has completed (success or failure).

Fields

tool_name: String

Name of the tool that was executed.

result: ToolResult

The outcome of the tool execution.

wait_ms: u64

Time spent waiting for a concurrency permit (semaphore), in milliseconds.

execute_duration_ms: u64

Wall-clock time taken to execute the tool after acquiring the permit, in milliseconds.

display_name: String

Human-readable tool name (prefix stripped, title-cased).

duration_display: String

Human-readable duration (e.g. “125ms”, “1.5s”).

IterationComplete

One full LLM→tool-execution cycle has completed.

Fields

iteration: usize

The 1-based iteration index that just finished.

tool_calls: usize

Number of tool calls executed during this iteration.

FinalAnswer

The loop has concluded and produced a definitive answer.

Fields

content: String

The complete final response text.

PromptProgress

Prompt-processing progress from the LLM backend.

Emitted during the pre-fill phase when llama-server is streaming prompt_progress frames. Surfaces token-level progress so the UI can show “processing prompt: 2048 / 8192 tokens”.

Fields

processed: u32

Number of tokens processed so far.

total: u32

Total number of tokens in the prompt.

cached: u32

Number of tokens served from KV cache (already processed).

time_ms: u64

Elapsed wall-clock time in milliseconds since processing began.

SystemWarning

A non-fatal system-level warning surfaced by the loop itself.

Unlike AgentEvent::Error, a SystemWarning does not terminate the loop — it informs the user that the loop encountered a recoverable condition (e.g. the model requested more parallel tool calls than the configured limit, and the loop is auto-retrying with a synthetic error fed back to the model).

suggested_action, when present, contains an actionable hint the UI can render verbatim (e.g. a CLI command to permanently raise a limit).

Fields

message: String

Human-readable description of the recoverable condition.

suggested_action: Option<String>

Optional actionable hint (e.g. CLI command) the UI can show to the user to permanently address the cause.

Error

A fatal error has terminated the loop.

Fields

message: String

Human-readable description of the failure.

Trait Implementations

impl Clone for AgentEvent

fn clone(&self) -> AgentEvent

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for AgentEvent

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

Formats the value using the given formatter.

impl Serialize for AgentEvent

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.