37 tok/s from Qwen3-Next-80B (48 GB) on a 32 GB laptop, 24 without drafts vs 2.8 for llama.cpp; reading 62 vs 16 tok/s. Faster than llama.cpp on all 13 models compared in Q4_K_M. LLMs bigger than your RAM, pure C, CPU or GPU.
C avx2ccpu-inferenceggufinference-enginelinuxllmlocal-llmmixture-of-expertsmoenvmequantizationqwen3simdspeculative-decodingvulkan
/* SPDX-License-Identifier: GPL-3.0-or-later
Copyright (C) 2026 Maurizio Cammalleri */
/*
* https_get.h - a GET over HTTPS (https_get.c), for Janas's services.
*/
#ifndef JANAS_COMMON_HTTPS_GET_H
#define JANAS_COMMON_HTTPS_GET_H
#include <stddef.h>
#include "llm/json.h"
struct janas_net_policy;
/* GET of url (https), following redirects, the body into out (with a 0
after it, not counted). The HTTP status (200...), or -1 with the reason
in err. */
int janas_https_get(const char *url, const char *const *headers,
size_t n_headers, struct janas_buf *out, char *err,
size_t err_len);
/* The same, waiting at most timeout_ms for each part of the answer
(janas_https_get: 15 s). */
int janas_https_get_ms(const char *url, const char *const *headers,
size_t n_headers, int timeout_ms, struct janas_buf *out,
char *err, size_t err_len);
/* The same for a URL someone else gave (http too): every address it and
its redirects resolve to checked against policy (common/mcp_policy.h),
the connection made only to one that passed. */
int janas_https_get_policy(const char *url,
const struct janas_net_policy *policy,
int timeout_ms, struct janas_buf *out, char *err,
size_t err_len);
#endif