Struct GraphTrace 

pub struct GraphTrace { /* private fields */ }

Opinionated tracing helpers to build a Graph-style hierarchy in LangSmith:

• Root run named Graph (RunType::Chain)
• Step runs under root (e.g. chatbot, should_continue, tools)
• Nested runs under steps (e.g. ChatOpenAI, tool runs)

This is intentionally generic: inputs/outputs are serde_json::Value so any app can provide LangSmith-compatible payloads such as { \"messages\": [...] }.

Implementations

impl GraphTrace

pub async fn start_root( inputs: Value, thread_id: Option<String>, ) -> Result<Self>

Starts the root Graph run (name: Graph, type: Chain) and POSTs it.

pub fn root_scope(&self) -> &RunScope

pub fn root_tracer(&self) -> &Tracer

pub async fn start_node_iteration( &self, node_name: &str, inputs: Value, ) -> Result<RunScope>

Starts a top-level step/node under the root run (POSTs it) and returns the scope. Use this for nodes like “chatbot”, “tools”, etc. that may contain nested runs.

pub async fn trace_llm_call( &self, parent_node: &RunScope, llm_name: &str, inputs: Value, outputs: Value, model_name: Option<&str>, ) -> Result<()>

Traces an LLM call within a parent node (e.g., within “chatbot”). llm_name is typically “ChatOpenAI” or similar. model_name is optional (e.g., “gpt-4o-mini”) and will be added to inputs if provided.

pub async fn trace_decision( &self, parent_node: &RunScope, decision_name: &str, inputs: Value, outputs: Value, ) -> Result<()>

Traces a routing/decision step (e.g., “should_continue”).

pub async fn trace_tool_call( &self, parent_node: &RunScope, tool_name: &str, inputs: Value, outputs: Value, ) -> Result<()>

Traces a tool call within a parent node (e.g., within “tools”). tool_name should be the actual tool name (e.g., “calculator”).

pub async fn end_root(self, outputs: Value) -> Result<()>

Ends the root run with the provided outputs (PATCH). Consumes self.