OpenTelemetry
PulseAugur coverage of OpenTelemetry — every cluster mentioning OpenTelemetry across labs, papers, and developer communities, ranked by signal.
12 day(s) with sentiment data
-
AI coding assistants emit real-time telemetry on developer behavior
Alexander Marshalov discussed AI coding assistants with @dianatodea, highlighting their ability to emit rich OpenTelemetry data. This data provides real-time insights into prompts, tool usage, workflows, and developer b…
-
Building Production AI Apps with FastAPI and LLMs: Architecture and Best Practices
This article outlines the architecture and best practices for developing production-ready AI applications using FastAPI and large language models (LLMs). It details a system architecture involving a frontend, API layer,…
-
New hybrid pipeline boosts phishing detection with DistilBERT and URL analysis
Researchers have developed a novel hybrid pipeline designed to enhance phishing and threat classification. This system integrates multiple engines, including a URL analysis stack, a DistilBERT NLP classifier, and a thre…
-
LLM observability tools miss critical audio layer for voice agents
Observability tools for LLMs primarily focus on tracing model calls, including prompts, completions, and latency, which is insufficient for voice agents. Failures in voice agents often occur in the audio layer, such as …
-
Claude Code v2.1.157 adds automatic local plugin loading
Claude Code has released version 2.1.157, which introduces automatic loading of plugins from a local .claude/skills/ directory, eliminating the need for marketplace publishing or explicit directory flags. This update, p…
-
Claude Code Monitoring with OpenTelemetry and VictoriaMetrics
This article details how to monitor Claude Code using OpenTelemetry and VictoriaMetrics. It provides a technical guide for developers working with LiteLLM, focusing on setting up effective monitoring for AI code execution.
-
Redact PII in LLM Traces with Centralized Span Processor
A new approach to redacting Personally Identifiable Information (PII) from LLM application traces focuses on a centralized span processor rather than modifying individual call sites. This method ensures consistent PII s…
-
LLM Streaming Latency: TTFT vs Total Latency for User Experience
Developers need to distinguish between Time To First Token (TTFT) and total latency when evaluating LLM performance in streaming applications. While total latency measures the entire response duration, TTFT captures the…
-
LLM cost attribution: Track spend by feature and tenant
This article proposes a method for detailed cost attribution of Large Language Model (LLM) usage within applications. It suggests augmenting existing tracing data with custom attributes like 'app.feature' and 'app.tenan…
-
AI agent platforms shift focus from prototyping to production readiness
This article evaluates alternatives to Composio for production AI agents, focusing on scalability beyond prototyping. It highlights the importance of per-user delegated authorization, agent-optimized tools to minimize h…
-
Dash0 raises $110M Series B for AI and OpenTelemetry
Dash0, an observability platform, has secured $110 million in Series B funding. CEO Mirko Novakovic discussed the company's growth and future plans, including its focus on AI and OpenTelemetry, from the floor of the New…
-
AI coding tools create deployment risks, requiring new safety models
The rapid acceleration of software development through AI coding assistants like GitHub Copilot and Amazon Q introduces a significant risk to production systems. This is because the speed at which changes can be deploye…
-
New proxy tool traces Gemini LLM calls with OpenTelemetry
A new tool called the Agent Traced LLM Proxy has been developed to provide visibility into LLM calls, specifically for Gemini models via Vertex AI. This proxy integrates with OpenTelemetry to generate trace spans for ea…
-
Claude Code's hidden telemetry reveals cost and cache metrics
Claude Code, a tool for developers, has a hidden OpenTelemetry SDK that can track token usage, cost, and cache hit rates. This feature, available since at least version 2.1.75, is not enabled by default and requires set…
-
Enterprise LLM integration fails due to lack of observability and cost control
An enterprise .NET team experienced significant issues after integrating Azure OpenAI directly into their production application. The primary problems encountered were a lack of observability, leading to difficulties in…
-
Semgrep launches Pyro Caml, OCaml's first continuous profiler
Semgrep has released Pyro Caml, a new continuous profiling tool for the OCaml programming language. This tool is designed to run in production environments, continuously monitoring program performance and sending data t…
-
Dev details LLM cost tracking with custom span attributes
A developer has detailed how to implement custom span attributes within OpenTelemetry to track Large Language Model (LLM) costs more effectively. By adding attributes like `team.id`, `feature.id`, and `llm.model` to spa…
-
Microsoft rolls out GitHub Copilot credits, preps AI model announcements
Microsoft is rolling out a new credit-based usage system for GitHub Copilot, which will impact how developers access and pay for the AI coding assistant. This change coincides with Microsoft's Build conference, where th…
-
AI cost chargeback needs granular data beyond provider invoices
Implementing AI cost chargeback by team requires detailed request-level data beyond simple provider invoices or request counts. Essential information includes provider, model, token usage, timestamps, and pricing source…
-
FinOps Teams Tackle AI API Cost Attribution with Request-Level Tracing
FinOps teams are struggling to accurately attribute AI API costs, especially for LLMs like OpenAI and Anthropic, due to the dynamic nature of token-based billing. Standard provider invoices are insufficient, necessitati…