Exporters

Send AgentCost data to your existing observability stack.

OpenTelemetry

Export trace data as OpenTelemetry spans to Datadog, Jaeger, Grafana Tempo, or any OTel-compatible backend.

Installation

pip install agentcostin[otel]

Setup

from agentcost.otel import install_otel_exporter

# OTLP gRPC (Jaeger, Grafana Tempo)
install_otel_exporter(endpoint="http://localhost:4317")

# OTLP HTTP (Datadog)
install_otel_exporter(
    endpoint="https://trace.agent.datadoghq.com",
    headers={"DD-API-KEY": "your-key"},
    protocol="http",
)

What Gets Exported

Each trace becomes an OTel span with attributes:

agentcost.model — Model name
agentcost.provider — Provider name
agentcost.project — Project name
agentcost.cost — Cost in USD
agentcost.input_tokens — Input token count
agentcost.output_tokens — Output token count
agentcost.latency_ms — Latency in milliseconds
agentcost.status — success/error

Prometheus

Expose metrics for Prometheus scraping and Grafana dashboards.

Setup

The /metrics endpoint is available automatically when the server is running:

agentcost dashboard
# Metrics at http://localhost:8500/metrics

Available Metrics

Metric	Type	Description
`agentcost_trace_total`	Counter	Total trace events
`agentcost_cost_total`	Counter	Total cost in USD
`agentcost_tokens_total`	Counter	Total tokens (input + output)
`agentcost_latency_ms`	Histogram	Request latency distribution
`agentcost_errors_total`	Counter	Total error events

Labels: model, provider, project, status

Grafana Dashboard

Import the included Grafana dashboard:

Open Grafana → Dashboards → Import
Upload examples/grafana-dashboard.json
Select your Prometheus data source

Prometheus Config

# prometheus.yml
scrape_configs:
  - job_name: "agentcost"
    scrape_interval: 15s
    static_configs:
      - targets: ["localhost:8500"]
    metrics_path: /metrics