repository

Open-source repository for a NanoGPT training speedrun, focused on getting a 124M model to target loss on 8x H100 GPUs in under 90 seconds. The README includes run commands, Docker instructions, and a detailed list of model and systems optimizations.

Deterministic retrieval-only Q&A system for documentation that avoids LLM calls at runtime. It uses Gemini embeddings, local Qdrant storage, and a CLI chatbot to answer questions with low latency and near-zero per-query cost.

Graft builds a semantic code graph for coding agents so tools like Claude Code and Cursor can understand a repository faster with less re-exploration. It ships as a CLI with agent wiring, graph generation, and visualization, and supports local or hosted model providers.

An unofficial MCP server for cosmos.so that lets AI clients search, browse, and save visual moodboards. The repo includes install steps, authentication guidance, and concrete examples for Claude Code, Claude Desktop, and Cursor.

A local CLI for replaying what AI coding agents accessed on your machine, including file reads, secrets that entered context, and possible read-then-network-call exposure paths. It also scans exported ChatGPT, Claude, and Gemini histories for sensitive data.

An open-source project that runs full games of Diplomacy between 7 LLM-controlled powers, with negotiation rounds, validated orders, and saved transcripts for replay and analysis.

A zero-config CLI that scans repositories for AI-slop patterns like missing .env.example files, committed secrets, large files, duplicated code, TODO drift, and missing tests, then returns a Vibe Score from 0 to 100.

Local-first code intelligence for Claude Code that adds structural graph search, semantic memory, cross-project links, MCP integration, and automated hooks for coding workflows.

Hivemind is a shared brain for coding agents that captures traces, turns repeated patterns into reusable skills, and syncs them across a team. The repo shows installation commands, supported assistants, commands, and troubleshooting notes.

Open-source CLI and agent infrastructure for handling paid API requests. It detects 402 payment challenges, signs stablecoin payments with a local wallet, supports MCP, and includes a payment debugger plus catalog tooling.

Amnitex is a local, lossless memory layer for MCP-capable AI coding assistants. It installs as a CLI, exposes tools like search, recall, remember, and list keys, and is designed to work across clients such as Claude Desktop, Cursor, Cline, Continue, and Zed.

An open protocol for giving AI agents structured, real-time perception from external sensors and data sources. The repo includes architecture notes, quick-start install flows, and a sensor catalog concept.

ModelPeer is an open-source tool for cross-model peer review in coding workflows. It lets Claude Code, Codex CLI, and Gemini CLI review one another as read-only engineering peers, with commands for consultation, diff review, setup, updates, and safety controls.

A CLI and Python harness that lets AI agents control a real iPhone through macOS iPhone Mirroring or an Android phone over adb, with OCR, accessibility, and input primitives for automation.

A set of 20 Claude Code skills distilled from real engineering failures, each packaged as a short rule set with a pre-flight checklist and install instructions.

A command guard for AI coding agents that blocks destructive git and shell commands before they run, with support for multiple agent and CLI tools.

A cross-platform agent skills repository for consensus-based R&D and repo automation. It describes a multi-agent loop that generates independent plans, converges through a meta-judge, and supports platforms like Claude Code, Codex, Cursor, and Gemini.

LeanGuard is a Lean 4-based runtime guardrail system for agent tool use, with native temporal policies, Python integration, and optional MCP support.

WorldCut is a TypeScript package that checks whether observations from independent systems satisfy declared version and time relationships before an autonomous agent makes a decision.

A hands-on lab toolkit for Docker and container security, covering CIS checks, image hardening, signing, SBOMs, admission control, runtime security, and AI context-poisoning defenses.

Codecut is a CLI for keeping coding agents on track by turning project rules into automatic checks. It flags issues like missing tests, failing test output, temporary agent comments, long code comments, and large changes that need an explanation, then explains what to fix.

A self-healing knowledge layer for AI coding agents with trust-verified search results, session auto-sync, and filesystem-backed reconciliation.

An unofficial Grafana plugin for Hermes Agent that records LLM calls and tool executions as generations and sends OTEL traces and metrics to Grafana Cloud.

CLI and GitHub Action that scans repos, logs, and CI artifacts for leaked LLM reasoning-trace blocks from Anthropic, OpenAI, and Google.