Projects

The AI coding agent platform trusted by Fortune 500 teams. Blackbox AI lets you dispatch autonomous coding agents from your terminal, IDE, cloud, API, mobile, or browser. Multi-agent parallel execution with Chairman LLM evaluation — agents compete, collaborate, and ship code while you focus on what matters. Built with Next.js, real-time WebSocket streaming, and OpenAI-compatible API endpoints.

• Six-surface agent platform — CLI, IDE, Cloud, API, Mobile, Builder
• Chairman LLM multi-agent evaluation system
• Parallel agent dispatch with autonomous execution
• OpenAI-compatible API with real-time streaming
• Full-stack app generation from natural language
• Enterprise-grade security with E2E encryption

Stop re-teaching your AI coding agents the same rules. Memories.sh is a local-first memory layer that stores context in a SQLite database with vector embeddings, then generates native config files for every major coding tool — Cursor, Claude Code, Copilot, Windsurf, Gemini, and more. Three commands: store, recall, generate.

• Durable local state with semantic recall
• 100MB on-device embedding model — no API calls
• Native config generation for 13+ tools
• Scoped memory (global vs. per-repo)
• Export to JSON/YAML anytime — zero lock-in
• Built-in MCP server for browser-based agents

A chat-first creative studio where you describe what you want and AI builds it. Videos, images, memes, thumbnails, sprites — all generated through conversation. Includes a full suite of free cloud-powered utilities for video conversion, compression, background removal, and audio processing.

• AI-powered video editor with natural language control
• Chat-based Lab for end-to-end creation
• Image-to-video and text-to-video generation
• YouTube clip extraction and thumbnail generation
• Meme and social ad creator
• 15+ free utilities — no watermarks

An agentic design canvas — describe what you want to create, and the AI agent builds the entire workflow on a node-based canvas. Extract color palettes, reverse-engineer animations, generate React components from screenshots, create brand kits from a logo. Powered by frontier models from Anthropic, OpenAI, and Google.

• Conversational workflow builder
• Color extraction and typography matching
• Component analysis and code generation
• Animation capture to CSS keyframes
• Brand kit and design system starter
• 8 prebuilt workflow templates

One search across 9 icon libraries — Lucide, Phosphor, Heroicons, Tabler, Feather, Remix, Hugeicons, Simple Icons, and Iconoir. Get React, Vue, Svelte, or raw SVG output. Available as an MCP server for AI coding agents, a CLI for your terminal, and a web UI for browsing. Normalize stroke widths across libraries for visual consistency.

• 9 icon libraries in one unified search
• React, Vue, Svelte, and SVG output
• MCP server for Cursor, v0, and Claude
• Starter packs (shadcn-ui, dashboard, ecommerce)
• Stroke normalization across libraries
• Bundle output — single file, 70% smaller

A modern headless CMS built for teams who want the flexibility of a database with the approachability of a spreadsheet. Visual schema builder, real-time collaboration, and a type-safe API layer. Currently in active development.

• Visual schema builder
• Spreadsheet-like content editing
• Type-safe auto-generated API
• Real-time collaboration
• Role-based access control
• Webhook integrations

A recreation of Robert Axelrod's 1984 Prisoner's Dilemma tournament — but with modern LLMs instead of hand-coded strategies. Each AI model plays against every other model in repeated games with hidden round counts. We test both explicit game-theory framing and cloaked real-world business scenarios to see if framing changes cooperation rates.

• Cross-model tournament (8+ LLMs)
• 4,200+ games played
• Overt and cloaked prompt framings
• Hidden game length — no end-game exploitation
• Live leaderboard with cooperation rates
• Full experiment data downloadable