Skip to main content

Observability Overview

This guide covers the observability stack for AI/Run CodeMie — the tools and configuration options that give administrators visibility into platform health, application logs, and AI model usage.

Observability Stack

AI/Run CodeMie uses two complementary observability layers:

LayerToolsPurpose
Infrastructure logsFluent Bit, Elasticsearch, KibanaCollect, store, and visualize application and service logs
AI traces & metricsLangfuseTrack LLM requests, latency, costs, and conversation metrics

Infrastructure Logging

Log Collection — Fluent Bit

Fluent Bit runs as a DaemonSet on every cluster node and forwards container logs to Elasticsearch. It is deployed as part of the observability components during platform installation.

Log Storage — Elasticsearch

Application logs and conversation metrics are stored in Elasticsearch under two indexes controlled by environment variables in your CodeMie API deployment:

VariableDefaultDescription
ELASTIC_LOGS_INDEXlogs-codemie-infra*Index for application and API logs
CONVERSATIONS_METRICS_INDEXcodemie-conversationsIndex for conversation-level metrics

Log Visualization — Kibana

Kibana provides a pre-configured dashboard for browsing CodeMie logs and conversation metrics. Access it at the Kibana URL exposed by your deployment.

Log Level

Control the verbosity of application logs via the LOG_LEVEL environment variable in your CodeMie API configuration:

ValueDescription
DEBUGMaximum verbosity — includes request/response details
INFOStandard operational events (default)
WARNINGWarnings and errors only
ERRORErrors only

AI Traces & Metrics — Langfuse

Prerequisite: Assistants Evaluation extension

Langfuse is deployed as part of the optional Assistants Evaluation extension. You must deploy this extension before configuring Langfuse environment variables or enabling trace collection.

Langfuse captures detailed traces for every LLM call made through the platform. This enables per-request inspection of prompts, completions, latency, and token usage.

Configuration

Langfuse integration is configured via environment variables in the CodeMie API deployment:

VariableDescription
LANGFUSE_TRACESEnable or disable trace collection (true / false)
LANGFUSE_PUBLIC_KEYLangfuse project public key
LANGFUSE_SECRET_KEYLangfuse project secret key
LANGFUSE_HOSTURL of your Langfuse instance
LANGFUSE_BLOCKED_INSTRUMENTATION_SCOPESComma-separated list of scopes to exclude from tracing

What Langfuse Tracks

  • LLM request and response content
  • Token usage and estimated cost per request
  • Latency by model and endpoint
  • Conversation-level aggregations