Status: [UNTESTED] Endpoints found: http://127.0.0.1:8787/v1/chat/completions Source: https://github.com/hujanpanada/sakelar # Sakelar **One local endpoint for all your LLM providers โ€” with every token counted.** Sakelar is a small, self-hosted proxy that sits between your tools (Pi, scripts, apps) and your LLM providers (OpenAI, Anthropic, DeepSeek, Ollama, any compatible endpoint). Your tools talk to Sakelar with a single local API key; Sakelar forwards each request to the right provider and records exactly how many tokens it used โ€” input, output, cache read, and cache write โ€” per client key, provider, and model. No config files, no cloud account, no telemetry: everything (providers, keys, usage history) lives in one local SQLite database and is managed from a built-in web dashboard. ![Sakelar dashboard โ€” API token usage by provider, with recent requests](images/sakelar-dashboard.png) - ๐Ÿงฎ **Accurate token tracking** โ€” Today / weekly / monthly / yearly / all-time totals, with cache read/write separation and no double counting - ๐ŸŒ **Configurable timezone** โ€” default GMT+8 (`Asia/Kuala_Lumpur`) or auto-detected from browser, ensuring daily boundaries align with your schedule - ๐Ÿ”‘ **Multiple named client API keys** โ€” one per device or app (`sk-sk-โ€ฆ`), revocable without touching upstream provider keys - ๐Ÿ”Œ **Three API types** โ€” `openai-completions`, `openai-responses`, and `anthropic-messages` - ๐Ÿ“ก **Transparent streaming** โ€” tokens counted on the fly without buffering; client disconnects safely abort upstream requests - ๐Ÿงช **Live test harness & capability probing** โ€” test image inputs and reasoning effort levels directly from the dashboard - ๐Ÿชถ **Zero runtime dependencies** โ€” built purely on Node.js built-ins (`http`, `sqlite`, `crypto`) with a Neubrutalist HTML/CSS/JS dashboard (light & dark mode) > Sakelar deliberately does **not** estimate cost, translate between > protocols, or do fallback routing. It forwards same-protocol requests > transparently and counts what providers report. **Inspired by [Pi](https://pi.dev).** Sakelar was built by