Introducing Lit, Louie, and Dewey

Today we are releasing three new open-source tools from NERVOSYS, each written in Rust and built from the ground up for the age of AI agents. Together they cover the three layers where agents and software meet: version control, terminal interfaces, and graphical interfaces. Meet Lit, Louie, and Dewey.

For two decades, our tools have assumed a human is on the other end — reading freeform text, squinting at dashboards, clicking through dialogs. AI agents inherit all of that friction. They parse human-readable output with brittle regular expressions, work around interactive prompts, and guess at what a button does. Lit, Louie, and Dewey invert that assumption. Every one of them is agentic-first: structured data by default, a machine-readable ontology for discovery, and a human-friendly mode available with a single flag.

Lit — Version Control Built for Agents

Git was designed in 2005 for human developers at a terminal. Every interface — output formatting, error messages, interactive prompts, conflict markers — assumes a human is reading and responding. Agents must scrape Git's freeform text with fragile regex, dodge interactive prompts, and translate <<<<<<< conflict markers into something actionable.

Lit inverts this. It is a complete Git replacement written in Rust, with more than 65 commands and 30 Model Context Protocol tools. Every command emits structured JSON by default. Errors carry machine-actionable codes and remediation hints. Merge conflicts are structured objects, not text markers. Batch operations accept JSON Lines on standard input, and nothing ever stops to prompt for input. Humans get the same power through a single --human flag.

Lit is not limited to source code. Its pluggable content-type system versions CAD models, EDA schematics, manuscripts, databases, scientific datasets, media, and geospatial data — each with domain-appropriate diff, merge, and storage strategies. It ships with decentralized identity, capability-based delegation, agent trust scoring, swarm coordination with file leasing, and content-addressed peer-to-peer federation.

Security is foundational. Lit uses quantum-resistant cryptography — SHA3-512 and BLAKE3 hashing, ML-DSA-87 post-quantum signatures, and AES-256-GCM encryption at rest — with FIPS 140-3 self-tests at startup, sandboxed execution, per-IP rate limiting, and tamper-evident audit logs.

Install it today: cargo install litvc (the command-line tool is still named lit)

Louie — The TUI Framework for Agentic AI

Louie brings the same philosophy to the terminal. It combines the best of modern TUI frameworks — ratatui, bubbletea, ink — with a structured metadata layer that lets AI agents discover, inspect, and interact with every widget in your application.

For humans, Louie offers the Elm architecture, immediate-mode rendering, an animation system, and a rich widget set. For agents, every widget exposes its schema, capabilities, actions, and semantic role through a typed ontology. An agent connecting to a Louie app can ask: what widgets exist, what can I click, what fields accept input, what actions are available — and get structured JSON answers, with no hardcoded assumptions and no trial and error.

Add it to your project: cargo add louietui (the crate imports as louie)

Dewey — The GUI Framework for Agentic AI

Dewey does for graphical interfaces what Louie does for the terminal. It is a backend-agnostic Rust GUI framework with a pluggable Painter abstraction and optional GPU-accelerated rendering through egui and aga, our own standalone rendering backend. It provides a complete semantic ontology over every one of its 29 widgets.

AI agents can discover, inspect, and control graphical applications through a structured JSON Lines protocol — no screen-scraping, no pixel matching, no accessibility-tree hacking. Dewey ships with six rendering backends, from hardware-accelerated GPU to a headless test backend that records every draw call, so the same application runs on Windows, macOS, Linux, and the web, with or without a display. A WebSocket transport extends the same agent protocol to remote and cross-language control.

Add it to your project: cargo add deweygui (the crate imports as dewey, with its GPU backend available as aga)

aga — short for Agentic Graphics Acceleration — is a standalone agentic-first rendering backend and a drop-in replacement for wgpu. It ships separately as cargo add aga, so you can build hardware-accelerated, agent-discoverable graphics without pulling in the full Dewey framework.

Safe by Design

Releasing tools for agents means taking agent safety seriously. Across all three projects, every agent-facing entry point is hardened. Standard-input request loops cap how much they will buffer, so a malformed or hostile stream cannot exhaust memory before a size check runs. Servers enforce per-second rate limits and reject oversized requests. Output is compact and token-efficient by default, valid as JSON Lines, with pretty-printed formatting available only when a human asks for it. Operations that would normally block on a terminal fail fast with clear remediation instead of hanging an autonomous workflow.

Lit, Louie, and Dewey are available now on the NERVOSYS GitHub organization. We are building the foundation for a world where agents are first-class participants in software — and we are just getting started.