ChatGH

thedotmack / claude-mem

Persistent memory layer for AI coding agents: automatically captures every session's tool usage and actions, compresses them with AI into searchable structured observations (SQLite + hybrid vector search), and injects relevant context into future sessions so agents retain project knowledge across restarts. Works with Claude Code, OpenClaw, Codex, Gemini, Cursor, Copilot and more.

vercel-labs / agent-desktop

Computer use for AI agents: a Rust-built CLI and C library that reads and operates desktop apps on macOS, Windows and Linux via the OS accessibility tree. Snapshots return JSON with stable refs for every actionable element, so agents click, type and verify without pixel guessing — headless by default, with 56 commands and safe-to-retry error semantics.

max-sixty / worktrunk

Worktrunk is a CLI tool for Git worktree management optimized for parallel AI agent workflows. It makes worktrees as easy to use as branches, featuring automated hooks, LLM commit generation, build cache sharing, and a streamlined merge workflow.

JustVugg / colibri

Colibri is a lightweight, pure C inference engine with zero external dependencies for running frontier MoE models (like GLM-5, DeepSeek V4, and Kimi K3) on consumer hardware. By treating VRAM, RAM, and NVMe storage as a multitier memory hierarchy, it dynamically streams experts on demand to run giant models locally.

mattpocock / skills

A curated collection of agent skills for real engineering work — not vibe coding. Fixes the 4 biggest AI coding failure modes: misalignment (/grill-me), verbosity (shared language), broken code (/tdd), and architecture rot (/improve-codebase-architecture). Small, composable, model-agnostic, and battle-tested from decades of engineering experience.

p-e-w / heretic

Heretic is an automated tool for removing censorship and refusal behaviors from transformer language models without retraining. It combines directional ablation ('abliteration') with Optuna parameter optimization to suppress refusals while preserving model capabilities by minimizing KL divergence. Supports dense, MoE, and multimodal architectures, 4-bit quantization, and residual vector interpretability analysis.

floci-io / floci

Floci is a free, open-source local AWS emulator that runs 100+ AWS services (S3, DynamoDB, Lambda, SQS, and more) on your machine — no cloud account, auth token, or paid feature gates. Compatible with AWS SDKs, CLI, Terraform, and CDK, it's a drop-in LocalStack alternative: millisecond startup, tiny memory footprint, and real Docker-backed Lambda/RDS, ideal for local development, testing, and CI.

Usagi-org / ai-goofish-monitor

AI-powered Xianyu (Goofish) product monitoring built on Playwright: create tasks in natural language, watch new listings in real time or on a Cron schedule, and let a multimodal model analyze items with instant alerts via Telegram, Bark, WeChat Work and more. Features multi-task concurrency, keyword/price/region filters, account & proxy rotation, a full-featured Web UI, and one-click Docker deployment.