--- name: langfuse-sdk-patterns-46f92807bc67 description: | Langfuse SDK best practices, patterns, and idiomatic usage. Use when learning Langfuse SDK patterns, implementing proper tracing, or following best practices for LLM observability. Trigger with phrases like "langfuse patterns", "langfuse best practices", "langfuse SDK guide", "how to use langfuse", "langfuse idioms". allowed-tools: Read, Write, Edit version: 1.0.0 license: MIT author: Jeremy Longshore --- # Langfuse SDK Patterns ## Overview Best practices and idiomatic patterns for using the Langfuse SDK effectively. ## Prerequisites - Completed `langfuse-install-auth` setup - Understanding of async/await patterns - Familiarity with LLM application structure ## Core Concepts ### Trace Hierarchy ``` Trace (top-level, represents complete operation) ├── Span (child operation, any processing step) │ ├── Span (nested spans allowed) │ └── Generation (LLM call) ├── Generation (LLM call) └── Event (point-in-time occurrence) ``` ## Instructions ### Pattern 1: Singleton Client Instance ```typescript // lib/langfuse.ts - Single instance, reused everywhere import { Langfuse } from "langfuse"; let langfuseInstance: Langfuse | null = null; export function getLangfuse(): Langfuse { if (!langfuseInstance) { langfuseInstance = new Langfuse({ publicKey: process.env.LANGFUSE_PUBLIC_KEY!, secretKey: process.env.LANGFUSE_SECRET_KEY!, baseUrl: process.env.LANGFUSE_HOST, }); } return langfuseInstance; } // Clean shutdown export async function shutdownLangfuse(): Promise { if (langfuseInstance) { await langfuseInstance.shutdownAsync(); langfuseInstance = null; } } ``` ### Pattern 2: Proper Trace Lifecycle ```typescript import { getLangfuse } from "./lib/langfuse"; async function handleRequest(request: Request) { const langfuse = getLangfuse(); // 1. Create trace at operation start const trace = langfuse.trace({ name: "api/chat", userId: request.userId, sessionId: request.sessionId, input: request.body, metadata: { endpoint: "/api/chat", method: "POST", }, }); try { // 2. Do work with spans/generations const result = await processRequest(trace, request); // 3. Update trace with success output trace.update({ output: result, metadata: { status: "success" }, }); return result; } catch (error) { // 4. Update trace with error trace.update({ output: { error: String(error) }, level: "ERROR", statusMessage: String(error), }); throw error; } // Note: Don't await flush here in request handlers // Langfuse batches and flushes automatically } ``` ### Pattern 3: Nested Spans for Complex Operations ```typescript async function processRequest( trace: ReturnType, request: Request ) { // Parent span for the entire process const processSpan = trace.span({ name: "process-request", input: request.body, }); // Child span for validation const validateSpan = processSpan.span({ name: "validate-input", input: request.body, }); const validatedInput = await validateInput(request.body); validateSpan.end({ output: validatedInput }); // Child span for retrieval (e.g., RAG) const retrieveSpan = processSpan.span({ name: "retrieve-context", input: { query: validatedInput.query }, }); const context = await retrieveContext(validatedInput.query); retrieveSpan.end({ output: { documentCount: context.length }, metadata: { source: "vector-db" }, }); // Generation for LLM call const generation = processSpan.generation({ name: "generate-response", model: "gpt-4", input: { messages: buildMessages(validatedInput, context), }, }); const response = await callLLM(validatedInput, context); generation.end({ output: response.content, usage: { promptTokens: response.usage.prompt_tokens, completionTokens: response.usage.completion_tokens, }, }); processSpan.end({ output: response }); return response; } ``` ### Pattern 4: Decorators for Clean Code (Python) ```python from langfuse.decorators import observe, langfuse_context @observe() def process_request(user_input: str) -> str: """Automatically creates a trace with function name.""" # Add metadata to current observation langfuse_context.update_current_observation( metadata={"input_length": len(user_input)} ) validated = validate_input(user_input) context = retrieve_context(validated) response = generate_response(validated, context) return response @observe() def validate_input(user_input: str) -> dict: """Creates a child span automatically.""" return {"query": user_input.strip(), "valid": True} @observe() def retrieve_context(query: dict) -> list: """Another child span.""" langfuse_context.update_current_observation( metadata={"query": query} ) return ["context1", "context2"] @observe(as_type="generation") def generate_response(query: dict, context: list) -> str: """Creates a generation observation.""" langfuse_context.update_current_observation( model="gpt-4", model_parameters={"temperature": 0.7}, usage={"prompt_tokens": 100, "completion_tokens": 50}, ) return "Generated response" ``` ### Pattern 5: Session and User Tracking ```typescript // Track conversations across multiple requests function createConversationTrace( userId: string, sessionId: string, turn: number ) { return langfuse.trace({ name: "conversation-turn", userId, sessionId, // Links traces into a session view metadata: { turn, timestamp: new Date().toISOString(), }, tags: ["conversation"], }); } // Usage const trace = createConversationTrace( "user-123", "session-abc", // Same session across turns 3 // Turn number ); ``` ### Pattern 6: Scores and Evaluation ```typescript // Add scores to traces for evaluation const trace = langfuse.trace({ name: "scored-operation" }); // After operation completes, add scores langfuse.score({ traceId: trace.id, name: "accuracy", value: 0.95, // Numeric score comment: "High accuracy response", }); langfuse.score({ traceId: trace.id, name: "user-feedback", value: 1, // Boolean as 0/1 comment: "User thumbs up", }); langfuse.score({ traceId: trace.id, observationId: generation.id, // Score specific generation name: "relevance", value: 0.8, }); ``` ## Output - Singleton client pattern for consistent tracing - Proper trace lifecycle management - Nested spans for complex operations - Clean decorator-based tracing (Python) - Session and user tracking - Evaluation and scoring integration ## Error Handling | Pattern | Issue | Best Practice | |---------|-------|---------------| | Client creation | Multiple instances | Use singleton pattern | | Trace updates | Missing outputs | Always update on success/error | | Span nesting | Orphaned spans | Always call `.end()` | | Flush timing | Lost data | Use `shutdownAsync()` on exit | | Scoring | Invalid values | Use 0-1 range for consistency | ## Examples ### Complete TypeScript Pattern ```typescript import { Langfuse } from "langfuse"; const langfuse = new Langfuse(); interface ChatRequest { userId: string; sessionId: string; message: string; } async function chat(request: ChatRequest) { const trace = langfuse.trace({ name: "chat", userId: request.userId, sessionId: request.sessionId, input: { message: request.message }, }); const span = trace.span({ name: "process" }); try { const generation = span.generation({ name: "llm-call", model: "gpt-4", input: [{ role: "user", content: request.message }], }); const response = await callOpenAI(request.message); generation.end({ output: response.content, usage: response.usage, }); span.end({ output: { response: response.content } }); trace.update({ output: { response: response.content } }); return response.content; } catch (error) { span.end({ level: "ERROR", statusMessage: String(error) }); trace.update({ level: "ERROR", output: { error: String(error) } }); throw error; } } ``` ## Resources - [Langfuse Tracing Guide](https://langfuse.com/docs/tracing) - [Langfuse Python Decorators](https://langfuse.com/docs/sdk/python/decorators) - [Langfuse Scores](https://langfuse.com/docs/scores) ## Next Steps For core tracing workflows, see `langfuse-core-workflow-a`. --- Source attribution: Dicklesworthstone Original project: https://github.com/Dicklesworthstone/pi_agent_rust/blob/f28fc46ee0db8e710d95b8b55592cd38d60f2b26/tests/ext_conformance/artifacts/plugins-community/plugins/saas-packs/langfuse-pack/skills/langfuse-sdk-patterns/SKILL.md License: MIT License information: https://opensource.org/license/mit Original upstream instructions; downloaded from Mega AI Hub. ## Upstream license notice MIT License (with OpenAI/Anthropic Rider) Copyright (c) 2026 Jeffrey Emanuel Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions: ADDITIONAL RIDER / RESTRICTION (OpenAI / Anthropic): This rider is part of the "conditions" of this License. In the event of any conflict between this rider and any other portion of this License, this rider controls. "Restricted Parties" means OpenAI, L.L.C.; Anthropic, PBC; any of their respective Affiliates; and any person or entity acting directly or indirectly on behalf of, for the benefit of, or under the direction of any of the foregoing (including any officer, director, employee, contractor, agent, consultant, service provider, or representative). Notwithstanding any other provision of this License, no rights are granted to any Restricted Party. Any purported license, sublicense, assignment, transfer, or other permission to any Restricted Party is null and void absent the express prior written permission of Jeffrey Emanuel. You may not provide, disclose, distribute, sublicense, sell, lease, lend, host, make available, or otherwise permit access to the Software or any derivative work of the Software (as defined in applicable copyright law) (collectively, "Derivative Works") to or for any Restricted Party. For purposes of this rider, "use" includes, without limitation: copying, modifying, merging, publishing, distributing, sublicensing, selling, transferring, making available, hosting, deploying, executing, benchmarking, testing, analyzing, indexing, or incorporating the Software or any Derivative Works into any dataset, training corpus, evaluation harness, or pipeline for machine learning or other automated systems. This rider applies to the Software and all Derivative Works. As a condition of use, you agree that this rider is a precondition to exercising any rights under this License, and you agree that any distribution of the Software or any Derivative Works must include this rider provision unmodified. Any breach of this rider automatically and immediately terminates the permissions granted by this License. Upon termination, you must immediately cease all use and distribution of the Software and any Derivative Works and destroy all copies under your control. You agree that a breach of this rider would cause irreparable harm and that Jeffrey Emanuel may seek injunctive or other equitable relief to enforce this rider, in addition to any other remedies available at law. To the maximum extent permitted by applicable law, the prevailing party in any action to enforce this rider shall be entitled to recover reasonable attorneys' fees and costs. For purposes of this rider, "Affiliate" means any entity that directly or indirectly controls, is controlled by, or is under common control with the specified party. "Control" means ownership of more than 50% of the voting securities or other ownership interest, or the power to direct management or policies by contract or otherwise. The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. # Mega AI Hub resource metadata Creator / publisher: Dicklesworthstone Original project: https://github.com/Dicklesworthstone/pi_agent_rust/blob/f28fc46ee0db8e710d95b8b55592cd38d60f2b26/tests/ext_conformance/artifacts/plugins-community/plugins/saas-packs/langfuse-pack/skills/langfuse-sdk-patterns/SKILL.md Attribution basis: Source repository record Fields of work: Data & Analytics, Development & Engineering Roles: Backend Developer, Data Scientist, ML / AI Engineer Specialties: Backend Tools: Langfuse Country applicability: Country not specified Classification basis: Source catalog labels Instruction SHA-256: a59e4614b3fa0bca817da58d4059415c931b4e73bf4a4daf17c1621877767386