A catalog of experiments and products

Hyper-optimized, SIMD-accelerated CMA-ES implementation in Rust with Python bindings.

A high-performance evolution strategy optimizer. Features SIMD acceleration, Rayon parallelism, deterministic seeding, and a rich TUI. Designed for heavy optimization tasks where Python's GIL is a bottleneck.

Run high-performance optimization algorithms directly in the browser via WebAssembly.

Brings the power of the Rust CMA-ES engine to the web. Includes a visual interactive playground using Three.js and D3 to visualize convergence in real-time without server-side processing.

Headless browser that renders modern websites into clean, proven Markdown for agents.

Transforms complex, JavaScript-heavy web pages into structured Markdown. Uses deterministic Chrome for Testing to capture screenshots, perform OCR, and stitch everything into a format LLMs can actually read and cite.

Instant file search engine for Windows using NTFS USN journals and Tantivy.

A modern 'Everything' alternative written in Rust. Indexes metadata and content in real-time using low-level NTFS hooks. Features a memory-efficient background service and a GPUI-based frontend for instant results.

Intelligent resource guardrails to prevent dev tools from freezing your Linux desktop.

A comprehensive tuning script that wires together `ananicy-cpp`, `earlyoom`, and kernel tweaks. Prevents runaway builds or IDEs from locking up your UI by intelligently prioritizing interactive processes.

Unified MCP server exposing dozens of powerful tools to AI models.

The server counterpart to Ultimate MCP Client. Provides a comprehensive Model Context Protocol implementation that exposes a wide variety of tools (file operations, web scraping, database access, and more) through a unified interface. Designed for extensibility and seamless integration with Claude, GPT, and other frontier models.

Two-person rule for AI agents: peer review before dangerous commands execute.

Adds safety friction for autonomous agents. Three-tier risk classification (CRITICAL/DANGEROUS/CAUTION), cryptographic command binding with SHA256+HMAC, dynamic quorum based on active agents, and complete audit trails. Integrates with Claude Code hooks and MCP Agent Mail for notifications.

ACE-framework persistent memory for coding agents. Stores procedural, episodic, and semantic knowledge.

Implements the Autonomous Cognitive Entity framework to give agents human-like memory. Stores procedural knowledge (how-to playbooks), episodic memory (session histories), and semantic facts. Exposes MCP tools so agents can recall context across sessions without re-learning.

A high-performance 'Swiss Army Knife' FastAPI service for local LLMs.

Streamlines complex local LLM operations into a unified API. It handles multi-modal embedding generation, sophisticated semantic search, and grammar-enforced text completions with aggressive caching and RAM disk optimization for maximum efficiency.

Secure, browser-based tool to turn codebases into optimized LLM prompts.

A single-file HTML tool that runs entirely locally to protect your IP. It allows you to filter, combine, and minify code files from your machine into context-aware prompts for LLMs, streamlining the workflow for AI-assisted development without sending code to a third-party server.

Convert entire playlists into structured, searchable text with Whisper.

Automates the ingestion of massive amounts of video content, leveraging GPU-accelerated Whisper models to generate high-fidelity transcripts with timestamps. Includes an interactive HTML reader and NLP processing to turn hours of video into a queryable knowledge base.

Open-source Splunk alternative for efficient multi-server log analysis.

A cost-effective solution for aggregating and analyzing gigabytes of logs from distributed fleets. It uses aggressive parallelization to pull data, stores it in SQLite/Datasette for instant querying, and provides a lightweight web interface for deep insights without the enterprise price tag.

Lightning-fast Rust library for complex vector similarity metrics.

Delivers robust, high-performance computation of advanced similarity measures like Hoeffding's D and Kendall's Tau. Built in Rust with seamless Python bindings, it's essential for statistical analysis of large-scale embeddings where standard cosine similarity isn't enough.

Instantly turns SQLAlchemy ORM models into interactive SVG diagrams.

Eliminates the need to manually diagram database schemas. This utility parses your Python code and generates clear, directed graphs of your data models, making it effortless to visualize and communicate complex database architectures.

Seamless blogging platform using your GitHub repo as a CMS.

Transforms a standard GitHub repository of Markdown files into a high-performance, SEO-optimized blog. It combines the developer-friendly workflow of git-based content management with the speed and polish of a modern Next.js 14 application.

Arena for LLMs to compete and collaborate on coding challenges.

An automated framework where multiple LLMs iteratively refine solutions to complex problems. It fosters a 'survival of the fittest' evolution of code, using peer feedback to break out of local optima and produce solutions superior to any single model.

AI Cognitive Inoculation Protocol: Defense against prompt injection.

A security framework that 'inoculates' LLMs against semantic attacks. Inspired by cognitive psychology, it uses narrative directives and few-shot examples to help models recognize and neutralize sophisticated prompt injection attempts before they can execute.

Web-based therapeutic tool exploring 40Hz neural stimulation.

An accessible implementation of 40Hz gamma entrainment therapy. This web app delivers precise light and sound stimulation directly in the browser, democratizing access to promising non-invasive experimental treatments for Alzheimer's.

Discover who *really* matters among your stargazers.

Goes beyond vanity metrics to calculate an 'influencer score' for everyone who stars your repo. It helps maintainers identify key community members by analyzing the reach and activity of their stargazers.

Forensic data science tool for unmasking financial fraud.

A powerful analytical engine that applies network analysis and XGBoost to massive loan datasets. It successfully identifies suspicious patterns, synthetic identities, and coordinated fraud rings that slipped past standard banking checks.

Interactive 3D visualization for high-dimensional statistical testing.

Tackles the curse of dimensionality by projecting complex data onto random 3D subspaces. It fits ellipsoids and compares error distributions to provide a visual and statistical assessment of normality in high-dimensional datasets.

Adaptive, browser-based literacy tool for early childhood education.

A comprehensive single-file React application that teaches the alphabet through phonics, tracing, and gamification. It features adaptive difficulty and local progress tracking, making high-quality educational software accessible without backend infrastructure.

A vibe-coded reader for Henry Kissinger's 400-page undergraduate thesis.

Takes a painful scanned PDF and turns it into a clean, phone-friendly reading experience, with structure that plays nicely with modern models and human readers.

Web-based jazz chord progression editor with playback.

A web application that enables musicians to create, edit, and play jazz chord progressions. The tool combines intuitive interface design with advanced music theory algorithms for proper voicings and smooth voice leading. It runs entirely in-browser as a single HTML file with real-time audio playback.

Convert source code to LLM-ready prompts with real-time token counting.

Terminal app that transforms codebases into structured XML prompts for LLMs. Features Vim-style navigation, live tiktoken counting against 128K context windows, JS/CSS/HTML minification, .gitignore support, and a preset system for saved file selections. Cross-platform single binary.

Convert ChatGPT, Gemini, and Grok share links to Markdown and HTML.

CLI tool that extracts AI conversations from share links and saves them as clean, archivable Markdown and static HTML. Preserves fenced code blocks with syntax highlighting, creates deterministic filenames, and supports one-command GitHub Pages deployment. Uses Playwright for reliable extraction.

Battle-tested solutions for AI coding agents, terminal setup, and dev workflows.

Practical guides documenting real problems and their solutions. Covers destructive command protection for AI agents, host-aware terminal themes, persistent SSH sessions, macOS NFS auto-mount, budget 10GbE networking, Claude Code hook safety guards, and more. Copy-paste configurations for immediate use.

Keep dozens of GitHub repos in sync with a single command.

A beautiful, automation-friendly CLI for synchronizing GitHub repositories. Clone missing repos, pull updates, detect conflicts, and get actionable resolution commands. Pure Bash with git plumbing for reliable status detection, JSON output for scripting, and meaningful exit codes for CI.

Modular skills teaching Clawdbot to use the Flywheel toolkit.

A curated collection of Clawdbot skills for professional agentic coding workflows. Includes skills for NTM, Agent Mail, Beads Viewer, CASS, and more, plus cloud CLIs and workflow methodologies. Enables Claude-powered assistants across WhatsApp, Telegram, and web interfaces to orchestrate multi-agent development.

Complete skill management platform with MCP server for AI agent integration.

Store skills, search them, track their effectiveness, package them for sharing, and integrate them natively with AI agents via MCP server. Skills come from hand-written files, CASS mining, bundles, or guided workflows. Thompson sampling learns which skills work best. Multi-layer security: ACIP for prompt injection, DCG for command safety, path policy for traversal attacks, secret scanner for redaction.

Hook for Claude Code and Gemini CLI that blocks destructive commands.

High-performance safety hook that intercepts catastrophic commands before they execute. Protects against git reset --hard, rm -rf, DROP TABLE, and other irreversible operations that AI agents occasionally attempt. Works on Linux, macOS, and WSL with clear explanations and safer alternatives.

Curated prompt library for agentic coding workflows.

A Next.js platform for discovering, copying, and installing battle-tested prompts that supercharge AI coding agents. Browse categorized prompts, one-click copy, or install directly as Claude Code skills. Built with Next.js 16, React 19, and TypeScript.

Git-backed interactive issue list with dependency tracking.

A lightweight issue tracker that stores everything in git. Features dependency relationships between issues, markdown descriptions, and seamless integration with standard git workflows. Perfect for solo developers or small teams who want issue tracking without external services.

Manage multiple AI coding agent subscriptions and accounts.

Streamlines the management of multiple AI coding agent accounts (Claude, Codex, Gemini). Track subscription status, API keys, usage limits, and billing across providers. Essential for power users running multiple agent sessions.

Central API gateway for the Agentic Coding Flywheel ecosystem.

Unified entry point for Flywheel tools, providing authentication, rate limiting, and routing across the ecosystem. Enables seamless integration between NTM, Agent Mail, Beads, and other Flywheel components.

Fast cross-file search and replace tool for codebases.

A high-performance CLI for searching and replacing text across large codebases. Supports regex patterns, preview modes, and bulk operations with git-aware filtering. Optimized for use in agentic workflows.

Export Suno AI music to CD-quality audio formats.

Tool for downloading and converting Suno AI-generated music to high-quality audio formats suitable for CD burning or archival. Preserves metadata and supports batch processing.

High-performance AI coding agent CLI with sub-100ms startup and native SSE streaming.

A lightweight AI coding agent written in Rust with instant startup, a ~15MB binary, and 7 built-in tools. Features session branching, structured concurrency via asupersync, and beautiful terminal output via rich_rust. Designed for speed-critical agentic workflows.

Transparent compilation offloading for AI agents via Claude Code hooks.

Intercepts cargo and gcc builds via Claude Code PreToolUse hooks and routes them to remote workers. Features sub-millisecond routing decisions, smart worker load-balancing, project affinity caching, multi-agent deduplication, and fail-open design for zero-disruption fallback.

Iterative specification refinement using GPT Pro Extended Reasoning.

Multi-round AI review automation for architecture optimization. Bundles project documents, runs iterative GPT-powered review sessions with background processing, and tracks round history with git integration. Features session attach/detach and a beautiful gum-based TUI.

Bayesian zombie process detection and safe cleanup for dev environments.

Uses a four-class Bayesian inference model to identify abandoned and zombie processes with conservative defaults. Features transparent decision evidence, staged SIGTERM→SIGKILL, protected process lists, and cross-platform support. Never auto-kills without explicit confirmation.

Spec-first, cancel-correct, capability-secure async runtime for Rust.

A structured concurrency runtime with regions, cancel-correct protocols with bounded cleanup, and capability-based context. Features two-phase effects, a deterministic lab runtime for testing, and guarantees no orphan tasks. Used as the async foundation for pi_agent_rust and fastmcp_rust.

Reference Rust implementation of TOON, a token-optimized serialization format for LLMs.

Spec-first Rust port of Token-Optimized Object Notation. Features streaming decode, deterministic output, native binary with no Node.js dependency, and token-efficiency optimizations including key folding and delimiter support. 100% spec conformance.

Network observer for AI CLI processes with provider attribution and SQLite logging.

Tracks outbound connections from Claude Code, Codex CLI, and Gemini CLI with automatic provider-aware tagging (Anthropic, OpenAI, Google). Features descendant process tracking, live terminal stats, SQLite durable logging, and flexible JSON output.

Claude Code hook that re-reads AGENTS.md after context compaction.

A SessionStart hook for Claude Code that reminds the agent to re-read AGENTS.md after context compaction, preventing loss of project conventions during long coding sessions.

Backup and sync settings across AI coding agents.

Backs up and synchronizes configuration for Claude Code, Cursor, Codex, and other AI coding agents. Preserves settings, MCP configurations, and customizations across machines and reinstalls.

Monitoring and observability dashboard for AI coding agent fleets.

Comprehensive monitoring tool for AI coding agent fleets. Tracks agent health, resource usage, and work progress across multiple concurrent sessions.

Cross-platform CLI for tracking LLM provider usage across Codex, Claude, and Gemini.

Tracks API token consumption and costs across multiple LLM providers (Codex, Claude, Gemini). Provides usage reports, cost breakdowns, and trend analysis for managing AI coding budgets.

Two-tier hybrid local search engine with HNSW ANN, full-text search, and cross-encoder reranking.

A production-quality local search engine in Rust combining HNSW approximate nearest neighbor search with Tantivy full-text search and cross-encoder reranking for optimal relevance. Part of the FrankenSuite.

Clean-room Redis reimplementation in Rust targeting full drop-in protocol parity.

A from-scratch Redis implementation in Rust with a Deterministic Latency Replication Core (DLRC). Targets full Redis protocol compatibility as a memory-safe, single-binary alternative. Part of the FrankenSuite.

Full pandas API reimplementation in clean Rust with columnar storage and 1,500+ tests.

A clean-room reimplementation of the Python pandas API in Rust with columnar storage, 7 IO formats (CSV, JSON, JSONL, Parquet, Excel, SQL, Feather), and 1,500+ passing tests. Part of the FrankenSuite.

JAX transform semantics in Rust with JAXPR-like IR, 110 primitives, and full AD coverage.

JAX's functional transformation semantics reimplemented in Rust. Features a JAXPR-like intermediate representation, 110 math primitives, full automatic differentiation (VJP+JVP), and 834 oracle test fixtures for conformance. Part of the FrankenSuite.

Memory-safe NumPy reimplementation in Rust with 2,282 passing tests and bit-exact RNG parity.

A clean-room NumPy reimplementation in Rust across 9 crates with zero unsafe code. Achieves bit-exact RNG parity with NumPy and passes 2,282 conformance tests. Part of the FrankenSuite.

Clean-room SciPy reimplementation with Condition-Aware Solver Portfolio (CASP).

SciPy's scientific computing stack reimplemented in Rust with a novel Condition-Aware Solver Portfolio that selects optimal numerical algorithms based on problem structure. Part of the FrankenSuite.

PyTorch reimplementation in Rust with Deterministic Autograd Contract (DAC).

A clean-room PyTorch reimplementation in Rust targeting absolute numerical parity via a Deterministic Autograd Contract. Includes legacy oracle setup for conformance testing against PyTorch outputs. Part of the FrankenSuite.

Rust-first Mermaid-compatible diagram engine with 25+ diagram types and deterministic layout.

A Mermaid diagram engine reimplemented in Rust that parses all 25+ diagram types with deterministic layout, 200+ tests, and WASM-ready compilation. Part of the FrankenSuite.

NetworkX graph library drop-in replacement in Rust with Python bindings.

A Rust reimplementation of NetworkX that can be used as a standalone Rust library or as a high-performance NetworkX backend via Python bindings. Part of the FrankenSuite.

Clean-room Whisper speech recognition model in Rust.

OpenAI's Whisper speech recognition model reimplemented in Rust for high-performance audio transcription without Python dependencies. Part of the FrankenSuite.

Convert and resume AI sessions across Claude Code, Codex, Gemini, and other providers.

A CLI tool that converts agent session formats between providers, enabling you to start work in Claude Code and continue seamlessly in Codex or Gemini. Preserves context, tool calls, and reasoning across platform boundaries.

High-performance Rust rewrite of MCP Agent Mail with 12-crate workspace and FrankenSearch integration.

A ground-up Rust reimplementation of MCP Agent Mail with a 12-crate workspace. Features mail-like coordination with Git artifacts, SQLite indexing, FastMCP server, and advanced hybrid search via FrankenSearch integration.

Automated multi-cloud VPS benchmarking for finding the best price/performance ratio.

Automatically provisions, benchmarks, and compares VPS instances across cloud providers. Tests CPU, memory, disk I/O, and network performance to find the optimal compute for agentic coding workloads.