openai-compatible
13 repositories
在 dsh 里装上这个插件即可,无需登录、注册或填 API Key,就能使用包括 DeepSeek V4.1 Flash、Kimi K3 在内的前沿模型——完全免费,不限量。 All you do is install this plugin in dsh: no login, no sign-up, no API key — the frontier models are just there, DeepSeek V4.1 Flash and Kimi K3 among them. Completely free, with no usage cap.
A zero-config gateway unifying 23+ LLM providers behind one API — Bifrost is a Go-based AI gateway giving one OpenAI-compatible API across 23+ providers, with automatic failover, load balancing, semantic caching, virtual keys and budgets. It deploys in seconds via npx or Docker, includes a web UI for config and monitoring, and adds under 200µs overhead at 5k RPS—aimed at teams building reliable, multi-provider AI apps and agents.
A fast routing layer that cuts LLM costs without code changes — Weave Router sits between agentic apps and multiple LLM providers, picking the right model per prompt in under 50ms. Written in Go and OpenAI-compatible, it works with tools like Claude Code and Codex, letting teams cut inference costs 40-70% by just swapping an endpoint.
🤖 Open-source, local-first alternative to Grok Bot, OpenAI dots and Claude Duet. Your own AI bots on any model: GPT, Claude, Gemini, Grok, DeepSeek, Kimi, GLM and 70+ more through one Modelflare key. Reach them in the app, WeChat or Feishu; they remember you, work in the background, run commands and use your computer. Pay per use.
A faster local inference server for coding agents on Apple Silicon — Rapid-MLX is an Apache 2.0, OpenAI/Anthropic-compatible inference server built on MLX, aimed at coding agents needing reliable tool calling. Benchmarks show it decoding faster than Apple's mlx-lm, oMLX, and LM Studio on the same Qwen weights, making it worth a look for local-LLM Mac users.
Native MTP speculative decoding speeds up local LLMs on Apple Silicon — MTPLX runs Qwen 3.8 Flash Next, Qwen 3.8 27B and Ternary Bonsai 2 27B on Macs using the models' own multi-token prediction heads for exact, batched speculative decoding. It reports roughly 2x decode speed over plain decoding, with a 27B model fitting on 16 GB Macs, via an OpenAI/Anthropic-compatible local server for tools like OpenCode.
One gateway to proxy 50+ AI clients behind a single base URL — Agent2API is a Rust gateway that takes over local AI client logins and exposes them through one base URL with unified OpenAI/Anthropic/Images APIs. It adds multi-account routing, failover, check-in and balance reports, plus a plugin system for extra providers, aimed at anyone juggling many AI client accounts.
A self-hosted gateway that unifies 40+ LLM providers behind one API — Otari is a Mozilla AI project that lets you run your own OpenAI/Anthropic-compatible gateway in front of 40+ model providers. It adds virtual API keys, pre-spend budget checks, usage tracking, and local routing policies for failover, useful for teams wanting to own their AI infrastructure instead of relying on a hosted proxy.
A multi-model Muse: personal AI voice assistant on ESP32 - wake word or knob, any OpenAI-compatible brain (STT/chat/TTS mix-and-match), streaming replies, out of the box. Built on muse-gadget-sdk.
Capri — an AI desktop companion for Windows with local/API models, Hugging Face, optional voice, memory and permission-based desktop tools.
The lossless, headless coding agent harness. One 7 MB static binary, four tools, plain stdin/stdout, no silent cuts.
A single Go binary that wires any LLM to MCP servers and Skills — Shim is a zero-dependency CLI harness for coding agents: it connects any OpenAI-compatible or Anthropic-style model to MCP servers and Agent Skills, using a local mini-model to translate natural language into allowlisted commands. No Electron, Node, Python, Docker, or daemon—just a 7MB binary with plain JSON/JSONL state, aimed at terminal-first developers tired of per-tool setup friction.
One GPU read, many probability answers for bulk classification — spinf-serve reads content once and scores many narrow questions against it in a single forward pass, returning probabilities instead of generated text. Built for the authors' own large-scale news scoring job, it runs on a single 24GB GPU behind an OpenAI-compatible API, aimed at anyone doing bulk classification or content moderation rather than chat.