Module for recording and persisting per-model runtime telemetry locally.
// Source: https://github.com/vava-nessa/free-coding-models | License: MIT | Copyright 2025-2026 vava-nessa // Provided via justclaw.it.com; keep this notice with the code. /** * @file runtime-telemetry.js * @description Per-model runtime metrics captured by the daemon on every routed request. * * @details * 📖 Why this exists: * 📖 SWE-bench / tier scores are static and self-reported by providers — a model * 📖 can advertise 82% on SWE-bench and still fail 50% of real requests through a * 📖 flaky free-tier gateway. We track what *actually* happens on the wire: * 📖 - real success rate (success vs error calls) * 📖 - real avg tokens/second (throughput under our real traffic) * 📖 - recent calls (last 50, capped) for debugging * * 📖 Distinct from src/core/telemetry.js ... (truncated -- full source via MCP)
See the full source, get the GitHub permalink, and search for more like it.
Get a free API key