Havuz

Carves your agent's tool outputs down to what matters. Never cuts what doesn't come back.

24/7 AI-DJ radio in Go. Persistent ffmpeg→Icecast with live ducking, LLM banter, BYOK, editable skills. ~50MB RAM.

⚔️ Compare AI models side by side — same prompt, streamed answers, live code previews, speed, tokens and cost. Keys never leave your browser

Local-first research tool with explicit fetch decisions, source evidence, and cited reports. CLI and MCP interfaces. Python, MIT.

One command to install, heal & prove ruflo (claude-flow) + agentic-qe + RuvNet Brain — plus ambidextrous Claude/Codex/OpenCode host routing,

Optimize LLM agent harnesses through tunable Formulas to dynamically enhance the agent, and improve those Formulas with optimizers based on

Big models. Small Macs. Zero excuses.

ast-outline lets AI coding agents pull exactly the code context they need — a repo skeleton, a file outline, or one symbol — instead of read

Own your LLM's web search: a local search->fetch->rank pipeline that replaces hosted web-search tools. Measured: matches hosted accuracy at

Blazing fast Go port of gemini-web2api. Convert Google Gemini web into OpenAI-compatible API. Zero cost, single static binary.

MIT-licensed, backend-agnostic AI agent orchestration loop in Python: orchestrator→worker→reviewer as a deterministic harness. Drive coding-

Turn scattered knowledge, operational data, and history into source-linked context that your agents can inspect, explain, and reuse.

A local review board for work an agent wants a human to look at — agent publishes, you annotate in the browser, feedback flows back as JSON.

Production-ready, bleeding-edge, upstream-first vLLM Docker image for the NVIDIA DGX Spark (GB10 / sm_121a), verified against real models on

面向 AI 时代的数据可视化规范与工作流库 — specification & workflow library for AI-era data visualization

A terminal AI agent you can fully undo - every file and shell action is snapshotted and reversible. Model-agnostic (any LLM), and connects t

Keep your Claude Code / Codex 5h rate-limit windows back-to-back — auto-pings each provider the moment its window resets. | 在 5 小时限额窗口重置的瞬间

Open-source email infrastructure for AI agents.

Benchmark your local LLMs: speed, memory, and quality, in one command. TUI leaderboard for Ollama, LM Studio, llama.cpp, and vLLM.

MIT-licensed TypeScript runtime for AI-native game characters and worlds, with durable state, reliable actions, memory, tools, and Unity, Go

Production-grade Agent Skills from Evil Martians — drop-in expertise for your AI coding agent

Local-first CLI that reads your Spec-Driven Development artifacts and recommends the best value-for-money AI model to implement them. Powere

Kyros — The Memory OS for AI Agents Give your AI agents secure, self-correcting, persistent memory in 3 lines of code. Three memory types

ICA Lens: compact ICA-based interpretability tools for exploring LLM activations. Code release for the paper.

A desktop workspace for crafting vivid AI personas and exploring what emerges when different personalities fuse.

Android keyboard with local AI (Ollama, Whisper, MCP) or cloud (Gemini, Groq, OpenAI)

【ICML2026 Spotlight】 T2PO: Uncertainty-Guided Exploration Control for Stable Multi-Turn Agentic Reinforcement Learning

Multi-agent orchestration & workflow engine. Declarative YAML workflows, LLM coordinator with hub-and-spoke mailboxes, race-safe delivery. O

macOS AI 会议纪要助手，支持麦克风与系统音频录制、实时转录和选择预设提示词生成会议纪要。

AI-native video editor driven as a local MCP server. Linux + Windows: ingest, transcribe, edit graph, melt/ffmpeg render, QC.

Angles Code CLI — 2MB coding agent with 2GB powers

Bring your own key (BYOK) relay to enable lightweight frontend-only apps to use LLMs using the user's API keys without hitting CORS.

Build LLM systems you actually control. A free, open engineering book + course — from tokenization to serving your own models. Mechanisms, t

Rust-first, local-first multi-agent runtime with a desktop HUD, policy-gated tools, voice, and worktree-isolated coding agents.

Multi-Provider AI Gateway, "Because we have LiteLLM at home" No personal logs by design. Model autodiscovery, Failover groups, High availabi

📊 Automated hourly benchmarks for 20+ NVIDIA NIM models — interactive dashboard, zero infra, self-hostable. Open-source & community-driven.

346 reusable character attributes for AI agent personas — compose, measure, and port system-prompt personas. Build once. Keep personality ev

Karpathy-inspired coding-agent guidelines packaged for Claude Code, Codex, Cursor, Gemini, OpenCode, Aider, Copilot, OpenClaw, and any AGENT

Wiki built using Karpathy method containing information about TPU performance optimizations and hooking it up to autoresearch optimization e

Generalizing ARIS from 科研 to 研究 — same five-step loop (plan · draft · adversarial review · iterate · persist) applied beyond academic resear

Serve clean Markdown from your Next.js site to AI agents, crawlers, and LLMs. Humans get HTML, agents get clean Markdown of the same pages.

Neuroscience-inspired memory framework for AI agents

Open-Ended Long-Horizon Text Game Generation for Test-Time Continual Learning Agents

Local-first CLI video producer: cuts silence and fillers, word-timed captions, face-aware framing, and LLM-planned code-rendered graphics

AI 生成原生可编辑 PPTX 的 Skill：Claude Code/Codex/Cursor 一句话出 PPT，形状文字都能直接改｜AI PowerPoint generator (editable .pptx)

Can LLM/MARL agents coordinate in long-horizon, open-ended tasks? NeurIPS 2026 (Evaluations and Datasets).

Frontier-class LLM inference on a laptop CPU — gpt-oss:20b at ~110 tok/s on Apple M4 Max, 7.5x llama.cpp, no GPU. From-scratch NEON/SME kern

llama.cpp fork with TurboQuant quantization (turbo2/3/4) and TriAttention GPU-accelerated KV cache pruning. 75 tok/s on Qwen3-8B / RTX 3080.

Semantic retrieval for AI agents: high-recall memory and code search that cuts coding-agent token usage by nearly 50%

An AI coding harness that shaped itself - repo awareness, surgical edits, validation, recovery, and safe diff approvals.