Maple LiteLLM Agent Tracing
Instruments LiteLLM agents and proxies with OpenTelemetry to track conversation sessions, tool invocations, token usage, and costs in Maple.
The Maple LiteLLM Agent Tracing skill guides developers through integrating Maple's OpenTelemetry observability platform with LiteLLM-powered AI agents and self-hosted LiteLLM proxies. It establishes end-to-end tracing by combining LiteLLM's v2 OpenTelemetry exporter with custom application spans, organizing every multi-turn conversation into a unified Maple Agent Session. This setup enables real-time tracking of complete chat transcripts, nested tool executions, token consumption, error diagnostics, and model spend across both Python SDK and proxy deployment environments.
主な機能
01Configures LiteLLM OpenTelemetry v2 loggers to map conversations into unified Maple sessions
02Instruments agent execution loops and tool calls with contextual invoke_agent and execute_tool spans
03Supports both native Python LiteLLM SDK calls and self-hosted LiteLLM Docker proxies
04Handles OpenTelemetry context propagation across distributed multi-agent workflows
05Tracks detailed token consumption, model costs, and execution latencies per agent turn
061,925 GitHub stars
ユースケース
01Monitoring multi-turn AI customer support agents built with LiteLLM for cost and latency optimization
02Debugging agent loop failures and tool execution errors in production AI agent applications
03Capturing full chat transcripts and tool execution metrics in self-hosted LiteLLM proxy deployments