See what you're actually spending on AI.
One CLI to track API costs, subscriptions, MCP tools, and self-hosted GPU infrastructure — with real TCO math, not estimates.
The CLI is live. Install it now.
Spendlayer CLI connects to your provider billing APIs, tracks subscriptions and MCP tool credits, scrapes your vLLM Prometheus metrics, and gives you one unified report — all from your terminal. No SaaS, no account required.
Install
$ go install gitlab.com/spendlayer/cli/cmd/spendlayer@latestOr download a binary from the releases page.
What it looks like
Spend Summary — Last 7 Days ┌────────────┬──────────────────────┬──────────────┬───────────────┬──────────┬────────────┐ │ Provider │ Model │ Input Tokens │ Output Tokens │ Requests │ Cost (USD) │ ├────────────┼──────────────────────┼──────────────┼───────────────┼──────────┼────────────┤ │ anthropic │ claude-sonnet-4 │ 42.3M │ 8.1M │ 1,847 │ $248.37 │ │ anthropic │ claude-haiku-3.5 │ 18.6M │ 3.2M │ 4,291 │ $21.80 │ │ openai │ gpt-4o │ 5.2M │ 1.4M │ 892 │ $27.00 │ │ openrouter │ deepseek/deepseek-r1 │ 12.1M │ 4.8M │ 2,103 │ $15.44 │ │ │ TOTAL │ 78.2M │ 17.5M │ 9,133 │ $312.61 │ └────────────┴──────────────────────┴──────────────┴───────────────┴──────────┴────────────┘
The self-hosted story
Spendlayer is the only tool that answers: “Are my GPUs running Llama at $1.76/MTok actually cheaper than Claude API at $15/MTok?”
If you run vLLM on your own hardware, spendlayer vllm compare calculates your real effective $/MTok from hardware amortization, electricity, and overhead — then compares it to cloud API pricing.
Self-Hosted Report — RTX 6000 Ada Workstation (30d)
═══════════════════════════════════════════════════════
Monthly TCO: $396.17/mo
Hardware (amortized): $291.67
Electricity: $54.50
Overhead: $50.00
Effective Cost
───────────────────────────────────────────────────
Cost per MTok (blended): $1.76
Cost per MTok (prompt): $0.88
Cost per MTok (generation): $3.52
Cloud Comparison
───────────────────────────────────────────────────
If this workload ran on cloud APIs:
OpenAI gpt-4o: $876.50/mo ← more expensive
Anthropic claude-sonnet: $1,176.60/mo ← more expensive
Verdict: Self-hosting at $396.17/mo is cheaper than all configured
cloud alternatives.What Spendlayer Tracks
API Spend
Pull billing data from Anthropic, OpenAI, and OpenRouter. Per-model breakdown with token counts, request volumes, and daily cost charts.
Subscriptions
Track Claude Pro, ChatGPT Plus, Cursor Pro, and any AI subscription. Auto-renewal detection, overflow tracking, and upgrade/downgrade recommendations.
MCP Tools
Monitor credit-based and per-call MCP tool servers. Burn rate projections, balance tracking, and auto-discovery from your Claude Desktop and Claude Code configs.
Self-Hosted GPUs
Scrape vLLM Prometheus metrics, calculate hardware TCO with real amortization math, and compare your effective $/MTok against cloud API pricing.
Connect your providers in minutes
Bring your API keys. We handle the rest.
Web Dashboard
We're building a web dashboard to complement the CLI with team features, historical analytics, and alerting.
Unified Team View
See every provider, every team member, every project in one dashboard. Real-time spend data with full attribution.
Anomaly Detection
Get notified before a spend spike becomes a budget crisis. Thresholds per team, project, or provider.
Historical Analytics
Trend analysis, budget tracking, and cost forecasting across all your AI spend categories.
Get early access to the web dashboard
The CLI is free and open source — install it now. The web dashboard is in development. Join the waitlist to get early access when it launches.