Home
Curated open-source AI projects with verified GitHub data and hands-on breakdowns.
h4ckf0r0day/obscura (19,954 stars, Rust, Apache-2.0, created 2026-04-13) is an open-source headless browser engine for AI agents and web scrapers, written in Rust with embedded V8, 30MB memory, 85ms page load, built-in anti-fingerprint stealth mode with 3,520-domain tracker blocking, CDP protocol for Puppeteer/Playwright compatibility, and a built-in MCP server for direct AI agent integration.
Andrew Ng's open-source desktop AI coworker (12,250 GitHub stars, Python, MIT, created 2026-07-20). Runs locally, delivers finished work not chat. BYO model with 12+ providers, 25+ connectors, approval-gated actions, scheduling.
esengine/DeepSeek-Reasonix (28,575 stars, 1,836 forks, Go, MIT, created 2026-04-21, pushed today) is a community-built DeepSeek-native terminal coding agent -- not an official DeepSeek product. It is tuned around DeepSeek's prefix cache: cache-hit input costs 0.02 yuan vs 1 yuan for misses, a 50x gap. A single static Go binary, config/plugin-driven (reasonix.toml), supporting dual-model executor+planner, MCP plugins, and cross-compilation to 6 platforms. Includes four install paths and peer comparison.
TencentCloud/CubeSandbox (10,818 stars, Rust, Apache 2.0, created 2026-04-10, pushed today) is Tencent's open-source secure sandbox service for AI agents, built on RustVMM+KVM hardware-level isolation, creating sandboxes in <60ms with <5MB memory, E2B SDK compatible, supporting single-node and K8s cluster deployment with AutoPause/AutoResume.
github/spec-kit (125K stars, Python, MIT, GitHub official) is a spec-driven development toolkit that codifies the spec->plan->tasks->implement four-phase gate into a runnable workflow, supporting 30+ agent integrations with an Extensions/Presets/Bundles three-layer extension system.
getagentseal/codeburn (9,092 stars, TypeScript, created 2026-04-13) is a free, local tool to track AI coding token usage and cost, a desktop app across macOS/Windows/Linux with 36 integrations showing where your AI spend goes.
XiaomiMiMo/MiMo-Code (12,598 stars, TypeScript, MIT, created 2026-06-10, Xiaomi official) is a terminal-native AI coding assistant positioning "models and agents co-evolve", with persistent memory across sessions, multi-agent (build/plan/compose), Claude Code import, and multi-LLM channels.
obra/superpowers (264,481 stars, MIT, Shell) isn't another coding agent -- it's a complete development methodology layered on top of your existing one. It makes your agent ask what you really want before writing code, breaks out a spec, generates a TDD/YAGNI/DRY implementation plan, then dispatches subagents to work through each task autonomously for hours. Supports 11 coding agents including Claude Code, Cursor, Codex, and Kimi Code.
LobeHub (formerly LobeChat) is an 80k-star open-source AI agent orchestration framework on GitHub, evolved from a "chatbox alternative" into a "Chief Agent Operator." The article breaks down its positioning shift, differences from OpenAI WebUI/Claude official, self-hosting routes and barriers, the 10,000+ MCP plugin ecosystem, and warns that its license is the LobeHub Community License (not MIT)-commercial derivatives require a paid license.
Strix is a 46k-star open-source AI penetration testing tool on GitHub, built in Python under Apache-2.0 by the usestrix org. Not a static scanner but an autonomous AI agent that runs code, attempts exploitation, and produces reproducible PoCs-like a real security researcher. Covers positioning, full pentesting toolkit with multi-agent orchestration, differences from traditional SAST/DAST, local CLI vs cloud platform, barriers and costs, and target users.