About Model Meter

The problem we're solving

Engineering teams building with LLMs share the same pain: costs are invisible until the invoice arrives, there is no attribution to tell which feature or team drove a spike, and provider dashboards don't break down by your projects or environments. The average team discovers a costly AI bug 3–4 weeks after it ships.

Our approach

One place to see all your AI spend — attributed, explainable, and under control. Model Meter is not an LLM proxy, not a prompt debugger, and not an evaluation platform. It is a focused cost visibility and spend attribution tool. The SDK instruments existing API calls via OpenTelemetry without sitting in the critical path. Your AI requests go directly to the provider.

• Not a proxy — SDK instruments calls via OTel, requests go directly to the provider
• Privacy by default — prompts and completions are never collected
• Idempotent ingestion — retries cannot inflate usage or cost
• Effective-dated pricing — price changes do not rewrite historical records
• Tenant isolation — PostgreSQL RLS enforces workspace boundaries at the DB layer

Built in the open

The source code, documentation, and contribution guide are at https://github.com/jayanti-prajapati/model-meter. Hosted plans will be paid; the source stays available.

• Source available on GitHub — inspect, modify, contribute
• SDK-first — two lines of code, no proxy in the call path
• Async telemetry — never blocks or slows your AI requests
• Contributions welcome: pricing fixtures, gateway adapters, dashboard improvements, docs