LLM observability
In plain wordsRecording and inspecting what an AI application actually did on each request — prompts sent, documents found, tools used, answers given, time and cost — so problems can be found and fixed.
The capture, storage and analysis of detailed telemetry from applications built on language modelslarge language model, n.: A computer program trained on enormous amounts of text to predict the words that come next. At vast scale, that simple skill lets it chat, answer questions, translate, summarize and write code., so that their behavior can be inspected, debugged and improved. The core unit is the trace: a tree of spans recording each model call (prompt, parameters, output, tokentoken, n.: The small chunk of text an AI language model reads and writes — often a word, part of a word or a punctuation mark. AI usage, limits and prices are counted in tokens. counts, latencylatency, n.: How long you wait between asking an AI system for something and getting the answer. Lower latency means a snappier, more responsive system., cost), retrieval step, tool calltool call, n.: A single request from an AI model to use one of its tools — naming the tool and filling in the details — which the surrounding software then carries out and reports back on., guardrail check and subagentsubagent, n.: A helper AI agent that a main agent starts to handle one piece of a job; it works in its own separate workspace and reports back a short summary when it is done. within one request or agent runagent loop, n.: The repeating cycle at the heart of an AI agent: the model decides what to do next, software carries it out, the result is shown to the model, and the cycle repeats until the task is finished.. Observability platforms add search and replay of traces, user feedback, automated scoring with model gradersLLM-as-a-judge, n.: Using one AI language model to grade or compare the answers of another, as a cheaper and faster stand-in for human reviewers., dashboards of cost and quality over time, clustering of failure types, and export of interesting traces into test setsgolden dataset, n.: A small, carefully checked set of example questions with known good answers, used as a trusted yardstick to test an AI system every time it changes., tying production to evals-driven developmentevals-driven development, n.: Building an AI product by first writing tests that show what good answers look like, then accepting changes to prompts, models or code only when the test scores improve.. Instrumentation increasingly follows OpenTelemetry's generative-AI semantic conventions. Tools include LangSmith, Langfuse, Arize Phoenix, Braintrust and Weights & Biases Weave, as well as features of general monitoring vendors. Because traces contain user data, retention limits, access control and redaction are part of the design.
LLM observability showed the agent was re-reading the same file eleven times per task.
Origin
Grew out of model monitoringmodel monitoring, n.: Continuously checking an AI model after it goes live — what data it receives, what it predicts and whether it is still right — so that problems are caught before they cause harm. and software observability, a control-theory term that site-reliability engineers adopted in the 2010s. Dedicated tracing products for prompt chains and agents appeared in 2023, among them LangSmith, Langfuse and Arize Phoenix, and OpenTelemetry began standardizing semantic conventions for generative-AI telemetry in 2024.
Usage note
Monitoring asks whether known metrics are healthy; observability aims to answer new questions after the fact from richly recorded data. For language-model systems the terms are often used interchangeably.
Related
- model monitoringmodel monitoring, n.: Continuously checking an AI model after it goes live — what data it receives, what it predicts and whether it is still right — so that problems are caught before they cause harm.
- LLMOpsLLMOps, n.: The everyday work of running chatbots and other products built on large language models: tracking prompts, testing answers, watching cost and quality, and updating safely when models change.
- evals-driven developmentevals-driven development, n.: Building an AI product by first writing tests that show what good answers look like, then accepting changes to prompts, models or code only when the test scores improve.
- prompt managementprompt management, n.: Treating the written instructions given to AI models like valuable code: storing them in one place, tracking every change, testing new versions and rolling back bad ones.
- AI gatewayAI gateway, n.: A checkpoint between an organization's apps and the AI models they use that handles access, costs, logging, safety checks and switching between providers in one place.
- agent loopagent loop, n.: The repeating cycle at the heart of an AI agent: the model decides what to do next, software carries it out, the result is shown to the model, and the cycle repeats until the task is finished.
- LLM-as-a-judgeLLM-as-a-judge, n.: Using one AI language model to grade or compare the answers of another, as a cheaper and faster stand-in for human reviewers.
Mentioned in
AI engineeringAI engineering, n.: The craft of building useful, reliable products on top of AI models — choosing models, writing prompts, connecting data and tools, testing quality and running it all safely at scale., AI stackAI stack, n.: The layers of technology an AI product depends on, from chips and data centers at the bottom, through the models, up to the apps people use, each built on the layer below., LangChainLangChain, prop. n.: A popular free software toolkit, and the company behind it, for building applications that connect AI language models to documents, databases, tools and multi-step workflows., model monitoringmodel monitoring, n.: Continuously checking an AI model after it goes live — what data it receives, what it predicts and whether it is still right — so that problems are caught before they cause harm., AI gatewayAI gateway, n.: A checkpoint between an organization's apps and the AI models they use that handles access, costs, logging, safety checks and switching between providers in one place., evals-driven developmentevals-driven development, n.: Building an AI product by first writing tests that show what good answers look like, then accepting changes to prompts, models or code only when the test scores improve., LLMOpsLLMOps, n.: The everyday work of running chatbots and other products built on large language models: tracking prompts, testing answers, watching cost and quality, and updating safely when models change., MLflowMLflow, prop. n.: A free software platform that records machine-learning experiments, packages the resulting models and keeps track of which versions are ready to be used., prompt managementprompt management, n.: Treating the written instructions given to AI models like valuable code: storing them in one place, tracking every change, testing new versions and rolling back bad ones., Weights & BiasesWeights & Biases, prop. n.: A company and online service that machine-learning teams use to log, chart and compare their training experiments, models and AI-app test results in shared dashboards.