v0.2.1

0.2.1 (2026-04-21)

Bug Fixes

• bump Chart.yaml to 0.2.0 and wire release-please to maintain it (#34) (59e9de0)

v0.2.0

0.2.0 (2026-04-21)

Features

• add team attribution, budget tracking, and alerting (M2) (#17) (c475fa9)
• add vLLM scraper with per-pod backend selection (#26) (30be439)
• FOCUS-compatible CSV export for UsageReport data (#31) (277423c)
• ship PodMonitor template for automatic Prometheus discovery (#24) (57e8eb8), closes #18
• surface DCGMReachable status condition on CostProfile (#28) (f44b538)
• unify cloud pricing on canonical YAML with refresh workflow (#29) (aed344a)

Bug Fixes

• bust Go Report Card badge cache (38ee787)
• stop UsageReport reconcile hot-loop on status updates (#25) (98ee49e)

Documentation

• add CostProfile sample library for common GPUs (#27) (8d9418d)
• batch-fill missing godoc comments (#32) (b92b57b)
• write launch runbooks (CNCF, OpenCost, HN/Reddit, FinOps) — not executed yet (#30) (4525822)

infercost-0.2.1

A Helm chart for InferCost - Kubernetes-native cost intelligence for on-premises AI inference

infercost-0.1.0

A Helm chart for InferCost - Kubernetes-native cost intelligence for on-premises AI inference

v0.1.1

Patch release: Docker images on GHCR, Homebrew formula update, OCI labels for repo linking.

v0.1.0

InferCost v0.1.0

The first release of InferCost: Kubernetes-native cost intelligence for on-premises AI inference.

Features

• CostProfile CRD: Declare GPU hardware economics (purchase price, amortization, electricity rate, PUE)
• UsageReport CRD: Auto-populated cost reports with per-model/namespace breakdown
• Cost engine: Scrapes DCGM for GPU power draw and llama.cpp/vLLM for token counts
• 12 Prometheus metrics: Cost-per-token, hourly cost, cloud comparison, GPU power, savings
• Cloud comparison: Verified pricing for 9 models across OpenAI, Anthropic, and Google
• CLI: infercost status and infercost compare with monthly projections
• REST API: /api/v1/costs/current, /api/v1/models, /api/v1/compare, /api/v1/status
• Grafana dashboard: 13 panels shipped as JSON
• Helm chart: Single-command installation with configurable DCGM endpoint and API server
• Honest results: Shows when cloud is cheaper than on-prem

Install

Homebrew (macOS):

brew install defilantech/tap/infercost

Helm:

helm install infercost infercost/infercost \
  --set dcgm.endpoint=http://nvidia-dcgm-exporter:9400/metrics

Binary: Download from the assets below.