{"name":"kapitsa","repo":"nace-ai/kapitsa","url":"https://midorreal.com/project/nace-ai-kapitsa","repository":"https://github.com/nace-ai/kapitsa","description":"Blazing fast Apple Silicon inference engine optimized for Qwen & Nemotron. OpenAI + Anthropic API compatible. Built by folks from nace.ai","facts":{"license":{"value":"Apache-2.0","source":"https://github.com/nace-ai/kapitsa","checked":"22 Nov 2026"},"language":{"value":"C++","source":"https://github.com/nace-ai/kapitsa","checked":"22 Nov 2026"},"homepage":{"value":"https://kapitsa.vercel.app","source":"https://kapitsa.vercel.app","checked":"22 Nov 2026"},"last_release":{"value":"v0.2.39, 20 Nov 2026","source":"https://github.com/nace-ai/kapitsa/releases","checked":"22 Nov 2026"},"releases":{"value":"50","source":"https://github.com/nace-ai/kapitsa/releases","checked":"21 Nov 2026"},"contributors":{"value":"68","source":"https://github.com/nace-ai/kapitsa/graphs/contributors","checked":"21 Nov 2026"},"archived":{"value":"no","source":"https://github.com/nace-ai/kapitsa","checked":"22 Nov 2026"}},"stars":851.0,"writeup":{"what_it_is":"Kapitsa is an inference engine for Apple silicon Macs that runs Qwen, Nemotron and Flint models locally. It serves OpenAI and Anthropic compatible APIs and can start coding agents such as opencode, Claude Code and Codex on a local model.","audience":"Mac users running local LLMs for coding agents","claims":[{"kind":"specific","claim":"Reports 351 tok/s decode for Qwen3.6 35B-A3B on an M5 Max","excerpt":"Kapitsa writes 351 tok/s on Qwen3.6 35B-A3B on an M5 Max","status":"not_checked"},{"kind":"specific","claim":"Requires a Mac with Apple silicon (M1 or newer) and macOS 14 or later","excerpt":"a Mac with Apple silicon (M1 or newer) and macOS 14 Sonoma or later","status":"not_checked"},{"kind":"specific","claim":"Has the fastest decode in 33 of 38 model and Mac pairs measured against nine engines","excerpt":"Kapitsa has the fastest decode in 33 of the 38 model and Mac pairs measured against nine engines","status":"not_checked"},{"kind":"specific","claim":"Can run a 180B model at 61 tok/s on a 48 GB M5 Pro","excerpt":"180B at 61 tok/s on a 48 GB M5 Pro","status":"not_checked"}],"alternatives":[],"written_by":"AI, from the project's README","date":"2026-11-21"},"why_now":null,"owner_supplied":null,"score":{"verdict":"real","hype":88,"reality":98,"gap":-10,"momentum":null,"confidence":60,"version":"score-2.0","rule":"verdict-2.0","calculated_at":"2026-11-22T08:00:00+00:00","verdict_source":"rule"},"early_read":null,"changes":[{"date":"2026-11-20T21:54:20+00:00","text":"Latest release v0.2.39","source":"https://github.com/nace-ai/kapitsa/releases"},{"date":"2026-10-18T08:49:31+00:00","text":"First release","source":"https://github.com/nace-ai/kapitsa/releases"},{"date":"2026-10-16T00:41:59+00:00","text":"Repository created","source":"https://github.com/nace-ai/kapitsa"}],"cite":"kapitsa is a GitHub project at https://github.com/nace-ai/kapitsa written mainly in C++, described there as \"Blazing fast Apple Silicon inference engine optimized for Qwen & Nemotron. OpenAI + Anthropic API compatible. Built by folks from nace.ai\".","attribution":"Data from Mid or Real, https://midorreal.com/project/nace-ai-kapitsa, checked 22 November 2026. Attribution with a link is required.","still_real":null}