qyl - AI Observability

WHAT YOU GET

OTLP

HTTP + gRPC ingestion on standard ports. Any language, any SDK. Point your existing OpenTelemetry setup at qyl and it works.

:5100/v1/traces · :4317 gRPC

GENAI

Token costs, model usage, latency percentiles. Native GenAI analytics, not bolted on. OTel Semantic Conventions v1.39.

OpenAI · Anthropic · Google · Azure · Bedrock · Cohere · Mistral

DASHBOARD

Embedded React UI with real-time SSE streaming. Sessions, traces, logs, GenAI analytics. Dark mode. Zero polling.

React 19 · Tailwind 4 · SSE live updates

MCP

30+ tools for Claude, GPT, and any AI agent. Your agents query their own telemetry. Sessions, errors, costs, traces.

qyl.list_sessions · qyl.get_genai_stats · qyl.analyze_session_errors

QUICK START

01

Run the container

docker run -d -p 5100:5100 -p 4317:4317 -v ~/.qyl:/data ghcr.io/ancplua/qyl:latest

02

Point your OTLP SDK at qyl

PYTHON

export OTEL_EXPORTER_OTLP_ENDPOINT=http://localhost:5100

.NET

builder.Services.AddOpenTelemetry()
    .WithTracing(t => t.AddOtlpExporter(o =>
        o.Endpoint = new Uri("http://localhost:5100/v1/traces")));

NODE.JS

OTEL_EXPORTER_OTLP_ENDPOINT=http://localhost:5100 node app.js

03

Open the dashboard

open http://localhost:5100

Your telemetry is there. Real-time. No setup.

GENAI-NATIVE

Other tools bolt GenAI on. qyl was built for it. Every LLM call is captured with full semantic context.

TOKENS

input · output · total

COST

USD per call · per session

LATENCY

P50 · P95 · P99

MODELS

usage · provider · tools

MCP FOR AI AGENTS

Your agents observe themselves. qyl exposes 30+ MCP tools so Claude, GPT, or any agent can query sessions, analyze errors, and track costs directly.

> qyl.get_genai_stats // token usage, costs, latency summary

> qyl.list_sessions // all AI sessions with error counts

> qyl.analyze_session_errors // every error in a session

> qyl.get_session_transcript // human-readable timeline

> qyl.search_spans // find anything in your telemetry

{
  "mcpServers": {
    "qyl": {
      "command": "dotnet",
      "args": ["run", "--project", "src/qyl.mcp"],
      "env": { "QYL_COLLECTOR_URL": "http://localhost:5100" }
    }
  }
}

VS. EVERYTHING ELSE

	QYL	JAEGER	GRAFANA	DATADOG
Setup	1 command	Cassandra/ES	Prometheus stack	SaaS account
Storage	DuckDB (1 file)	External DB	External DB	Cloud only
GenAI native	Yes	No	Add-on	Add-on
MCP tools	30+	No	No	No
Real-time	SSE streaming	Limited	Grafana Live	Yes
Cost	Free	Free	Free	$$$

CONFIGURATION

Variable	Default	Purpose
QYL_PORT	5100	HTTP API + Dashboard
QYL_GRPC_PORT	4317	gRPC OTLP ingestion
QYL_DATA_PATH	./qyl.duckdb	DuckDB file location
QYL_TOKEN	(none)	Auth token (disabled if unset)
QYL_MAX_RETENTION_DAYS	30	Days to retain telemetry
QYL_MAX_SPAN_COUNT	1,000,000	Max spans before cleanup