Open source frameworks and tools from our internal research stack. Architecture and methodology. Code available to select collaborators.
Research-grade tools, built in the open.
The following repositories document architecture, methodology, and interfaces from our internal research stack. We publish structure and specification — not implementation. Source code is available to select institutional collaborators.
A persistent agentic AI sidecar for developers. MU/TH/UR orchestrates multi-model AI workflows with a hierarchical governance layer, 5 automation systems that reduce token burn 50–70%, 16 native slash commands for Claude Code & OpenCode, a 6-file Soul Framework for persistent AI identity, and a native floating system health monitor. Personality blend: 50% MU/TH/UR · 25% TARS · 25% JARVIS.
A cyberpunk multi-agent AI command center built on OpenClaw. 16 specialist agents (MUTHUR orchestrator, SHIELD security monitor, finance-analyst, deep-research, and 12 more) coordinated via LangGraph/LangChain, secured by an always-on SHIELD security layer, accessible through a Next.js 16 dashboard with 13 pages. 100% local — Ollama inference, SQLite persistence, zero data egress.
A self-contained AI coding agent and data science harness built in Rust. WEYLAND orchestration engine, RALPH Loop execution (25 iterations), 5 specialist agents (Commander, Scout, Builder, Fixer, Oracle), MQ quality scoring, persistent Python DataView kernel (Plotly/Pandas), native Tauri v2 desktop app, full ratatui TUI, and security Containment Guard. Multi-provider: Anthropic, Ollama, xAI.
Institutional-grade alternative to Bloomberg Terminal. Five autonomous AI agents (Fundamental, Technical, Sentiment, Risk, Macro) on 63,606 survivorship-bias-free symbols. Options intelligence: 32 strategies, screener, backtester. Deep research swarm with Second Brain (LanceDB + 200+ books). Factor models, vectorbt, VIX surface, macro regimes. DuckDB 204M+ rows.
Additional repositories in preparation. Collaboration inquiries welcome.