Replace repetitive LLM calls with small, auditable ML models. Record → Shadow → Live, LLM kept as fallback.

Local code review hub for AI coding agents (Claude Code, Codex, Cursor, opencode). Agents submit with one CLI, you review files ranked by importance, feedback flows back. GitHub or plain git.

ReinAgent 是一个桌面 AI 编程客户端：把多个大模型接入你的本地工作区，用自然语言驱动它读写代码、执行命令、操作浏览器，并支持从手机远程查看与控制。

Switchya: a local coding workspace (desktop app + CLI) for the model providers you choose, powered by Switchyard Gateway, a self-hosted Rust LLM API gateway for OpenAI, Anthropic and Gemini protocols

Conversation tests for AI agents: a simulated user talks to your agent through its API, a judge checks every claim, and convy reports pass rate, tokens, time and what changed between builds.

Open-source marketing simulation for AI agents. Test ads, pricing, emails and pitches with synthetic buyers; inspect objections and plan real customer experiments. Self-hosted, MCP-ready, AGPL-3.0.

Coding-agent CLI with a code knowledge graph, memory and reviewable skills. CLI de agente de código com grafo de conhecimento, memória e skills revisáveis.

"Building Sarrera: Self-Hosted Enterprise AI Inference Gateway with RBAC, Token Quotas & Telemetry"

Open-source study studio in the spirit of NotebookLM: turn your sources into narrated decks, video and audio overviews, mind maps and notes, every claim cited.

麦门军师 —— 基于麦当劳 MCP 的点餐决策 Skill：自动领券、智能比价、营养把关、一句话点餐（麦当劳程序员创意开发大赛参赛作品）

帮被派去买全组午饭的那个人，30 秒完成一次不会有人不满意的点单。基于麦当劳 MCP 的多人点单 Skill。

给 AstrBot 的可落地 AI 记忆与生活层：五类记忆、混合检索、遗忘治理、情绪边界、自主生活循环。不覆盖人格。

One OpenAI-compatible endpoint over every free LLM API tier. Smartest model with quota left answers, then the next. Keyless first, proxy-aware, Persian docs, built for developers in Iran.

The decision layer for AI — typed decisions with confidence before your agent runs. Templates, Decision Schemas and API examples (cURL, Node.js, Python).

麦麦代送 mcd-send-a-meal：用麦当劳中国 MCP 给「别人」点一份（M-CODE 程序员创意开发大赛参赛项目）

把 Notion AI 网页登录态反代成本地 OpenAI / Anthropic / Responses 兼容接口：30+ 模型、Claude Code / Codex / cc-switch 接入、单文件 exe、自动改代码 agent。

my first, very vibecoded, cli coder, that soon will be very cool probably

Production-grade RLHF & post-training pipeline built from first principles in PyTorch. Features causal transformers, LoRA SFT, Bradley-Terry reward modeling, DPO, IPO, PPO with GAE, DeepSeek-R1 style GRPO reasoning, and automated statistical release gating.

Ask a question about a codebase, get the files that matter. Live, parallel file relevance with TypeSafe's Jev.

Voice-first macOS agent — drive your whole Mac by voice: sees the screen via Accessibility, acts via clicks/typing, verifies every step. Fast local System 1 + pluggable System 2 LLM. Beta.

Metodo completo de engenharia reversa para agentes de IA: 2 kits (app local e SaaS thin-client), 12 fases, portao de validacao proprio e o gate SEM-TRAVA que impede o agente de parar no meio. Instala em Claude Code, Codex, Hermes ou qualquer chat com shell. MIT

A terminal AI coding agent with a streaming TUI — multi-agent fan-out, LSP-powered fast inner loop, pull-based repo context, approval guardrails. One-line install on macOS / Linux / Windows. Rust + MIT, built on the phi-agent framework.

Local proxy that masks sensitive data between you and the LLM (Claude Code): the API only sees pseudonyms

Formally Verified AI Gateway

Agent Jobs: your main Claude Code session talks, decides and checks; everything longer than a few steps runs as a background claude -p job – own worktree, hard limits, guarded, one-line result. Cheaper context, free session.

Use OpenCode Zen's free models outside OpenCode — Pi extension with working tool calling and image input

TypeScript SDK for the three.ws v1 Agents API: create, start and chat with AI agents, run goals, automate, and trade from Solana agent wallets.

AI-powered video retrieval and clipping: transcribe with Whisper, segment with an LLM, query in natural language, export clips and SRT subtitles.

给 DeepSeek Harness 加一层心情。33 种心情各自持有数值，每 3~6 小时自己变 —— 它会累、会烦、会好奇；也会因为一轮里工具老报错而烦躁。附带：心情日记（每天 23:59 自己写）、跨会话记忆、品味库（判断的轨迹，跨会话带着看东西）、时事抓取（RSS / 网页 / 视频抽帧）、可选的主动开口。全部数据在本地、可搬到任意盘。MIT 开源。