Langfuse (YC W23)

Introducing Multi-Modal Support in Langfuse (YC W23) Tracing 💡 Comprehensive Analysis: Log and visualize text and image data as part of traces and observations for a richer, more detailed understanding of system behavior 🔍 Improved Context: Provides enhanced context for debugging and human-in-the-loop annotation We’re exploring building support across modalities — We’d love your feedback!

📊 New Add-on: Langfuse (YC W23), which provides product analytics for LLM apps, is now available as an add-on! 🤑 Spot Instances: When selecting node groups to deploy onto on AWS, spot instances are now an option (including GPU instance types). 💬 Slack Notifications Filtering: Notifications via Slack can now be filtered on an application level. 🔐 Enhanced Security: All Porter-managed infrastructure can now utilize an advanced service mesh. 🖥 Customizable GPU Resources: Users can now tailor their GKE clusters with flexible GPU options. More details in the changelog here 👉 https://lnkd.in/gXtwuchG

⚡️ OpenAI Structured Outputs now supported in Langfuse (YC W23) ❓ What are structured outputs in LLMs? 🧮 Structured outputs ensure that LLM responses consistently stick to a JSON schema. This simply means that you can rely on model outputs always following the same format (e.g. a number between 0-10, one out of four categories etc.). Structured outputs are a boon to increasing the reliability of LLM apps - they simply minimize LLMs notorious potential for going off the rails. Using structured outputs, developers can build apps that: • reliably extract structured data • fetch data & answer questions via function calling • create reliable multi-step workflows for LLMs to take actions 💡 In Langfuse you can easily track your OpenAI structured (and unstructured) outputs to improve your LLM application. 👇 Have a look at this cookbook to see how to trace structured outputs & supplied JSON schemas in Langfuse (in comments) 😱 Shoutout to Hassieb Pakzad for tweaking our OpenAI SDKs today

🤖🔍 AI Agent Observability! Everyone is talking about AI agents. You can easily set them up with no-code builders like FlowiseAI (YC S23), Dify, or Langflow, or use application frameworks like LangGraph (LangChain) or Llama Agents (LlamaIndex). ❓But what is AI Agent Observability? It makes your agents ready for production by tracking and analyzing performance and interactions. Langfuse (YC W23) provides real-time insights into metrics like latency, cost, and error rates, allowing developers to debug, optimize, and enhance their systems. 💡 Why it Matters: • Debugging & Edge Cases: Observability helps trace agent steps and test edge cases, crucial for resolving issues. • Balancing Accuracy & Costs: Monitoring model usage helps strike a balance between accuracy and costs. • Understanding User Interactions: Capturing user interaction data helps refine applications for better outputs. 👇 Link to article in comments if you want to learn more about AI Agent Observability

🚀 Exciting News! We are proud to share that Langfuse (YC W23) has been recognized as one of Germany's most promising AI startups by WirtschaftsWoche, one of Germany's leading business news publishers! 🌟 ⚡️ Standing out among 687 AI startups in Germany is an honor, as is being included among amazing companies like Tacto, Langdock, Parloa and Helsing. 🪢 A huge shoutout to our dedicated team, incredible users, and our open source community. 🔗 The full article is linked in the comments