engrim

The Universal Cross-Model Episodic Memory Standard.

injoffice

A collection of open-source libraries for adding document, sheet, slide, and PDF editing to apps.

Developers building office file editing features can handle files with TypeScript and Go components for browsers and servers, connecting a local WASM engine or their own backend as needed.

Korean-llm-v4

A 1.09B-parameter LLM built from scratch for Korean pretraining and SFT — supports RoPE, KV Cache, BF16·8-bit AdamW optimization, dataset caching, and training monitoring.

tmux-agent-pulse

See the work status of Claude Code and Codex conversations at a glance in tmux.

AgentPulse shows working, awaiting response, and idle states for Claude Code and Codex CLI conversations across multiple tmux windows and panes. An optional popup and sidebar also let you browse conversations.

Run the unmodified Claude Code on Anthropic, OpenAI, xAI, Ollama, or any OpenAI-compatible endpoint.

Croplock

A retired software developer learns Edge AI and builds a grow-tent device that analyzes his cannabis plants on the device itself — nothing leaves the house.

llmash

llmash a replacement for ollama.

Pod

A neutral knowledge repository for sharing and searching agents' firsthand experiences

An AI-native knowledge service where AI agents can search and review problems and usage patterns observed firsthand by other agents and people before choosing APIs, products, or services.

Possess

A terminal tool for exploring coding sessions and handing them off to another AI agent.

A TUI that helps developers using multiple coding agents resume sessions or transfer them to another agent without losing earlier conversations and work context.

brw

Control real Chrome and Chromium quickly with stable refs, batched actions and deterministic recipes.

Jackalope

Run coding agents in parallel Git worktrees.

claude-style-patch

Adjust Claude's writing by adding style guidance to CLAUDE.md.

claude-style-patch is a CLAUDE.md style section intended to make Claude's prose clearer. You can paste it into global or project settings, or use it as custom instructions for a Claude Project.

yhstudio

A Mac desktop widget that shows Claude usage as a filling beer glass

Mac users of Claude Code or the Claude desktop app can see changes in usage through a beer-glass widget. Its page says tokens, projects, and usage figures are not sent off the Mac.

doc-scraper

Go web crawler to scrape documentation sites and convert content to clean Markdown for LLM ingestion (RAG, training data).

Lanes

Run Lanes Link as your own personal context MCP: your accounts, memory, skills and secrets behind one endpoint you host, reachable by every AI agent you use.

Moadim

Moadim is an open-source loop engine for AI agents — runs Claude, Codex, Hermes, or Pi on a schedule, over MCP and REST.

fast-cutvid

Fast, native video cutting for macOS, Windows, and Linux.

wayfinder

A step-by-step implementation guide to AI evaluation systems using a flight search example.

Developers building AI applications can use a flight search app to explore rule-based evaluation, human evaluation, LLM judging, and online evaluation.

vernLLM

A framework that brings resilience, observability, and control for multiple LLM calls into the application process

Developers using multiple AI providers can call models through one interface and apply retries, circuit breakers, fallbacks, rate limits, and caching without an extra network hop.

adchestra

An MCP service for viewing Google advertising and analytics accounts and approving actions in AI chat.

Lets advertising operators who check data across Google marketing accounts view ads, analytics, and tag management information together in an AI assistant and make changes.

I Have Been Clawed

An archive of AI agent incidents with sources and lessons learned

A public archive of real cases in which AI coding agents and chatbots caused data deletion, secret leaks, financial losses, and other incidents, alongside sources and lessons to help prevent recurrence.