New open-source tool llmtee logs cost and latency of LLM API calls
Developer Garritfra released llmtee, a self-hosted proxy that sits between clients and OpenAI, Anthropic, or OpenAI-compatible APIs like OpenRouter, Ollama, vLLM and Groq, logging tokens, cost, latency and time to first token for every request while passing responses through unchanged. It runs as a single Go binary backed by SQLite or PostgreSQL, supports per-key daily and monthly budget caps that trigger HTTP 429 once exceeded, and exposes data via a web dashboard, CSV export, Prometheus metrics and an MCP server.
GoKawiil's interpretation of the reporting above, not reported fact.
The tool addresses a common pain point for developers and small teams: tracking and controlling spend across multiple LLM providers without building custom logging infrastructure. By requiring only a base URL change, it can be adopted with existing SDKs and tools like Claude Code or Codex without code modifications, which could lower the barrier to cost monitoring for smaller operations that lack dedicated gateway infrastructure. The creator notes it is explicitly not meant to compete with full gateway products offering routing or multi-tenancy, and remains in early 0.x development where configuration may still change.
- llmtee is a self-hosted proxy that logs tokens, cost, latency and time-to-first-token for LLM API calls without altering responses.
- It supports OpenAI, Anthropic and OpenAI-compatible providers, runs as a single Go binary with SQLite or PostgreSQL, and enforces per-key budget limits.
- It targets individuals and small teams rather than enterprise use, lacking routing, fallback or multi-tenant features, and is still in early 0.x release.
Source: github.com, 2026-11-11
Published there as: “Show HN: Log every LLM call with tokens, cost, latency, and time to first token”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.