Back to work
Personal product · R&D · In development · 2026

Argus Forge

What if an app could run a software project end-to-end? I'm building a desktop app that orchestrates a hierarchy of Claude Code agents — a CEO, supervisors, workers — like a small company running on your own machine. 100% local-first, zero SaaS.

Rust Tauri React 19 TypeScript Claude Code MCP Playwright
Challenge

A single AI agent gets lost on a full project: limited context, no memory, no quality control. What was needed was an organisation, not yet another agent.

Approach

A deterministic Rust layer (19 crates: locks, quotas, budget, git, events) framing hierarchical Claude Code agents. Truth lives in files — tickets, ADRs, schemas — not in a cloud.

Status

In active development — 100+ documented architecture decisions, 16 data schemas, Rust quality gates. The code reads, the product is being written.

01Context

A miniature company on your machine

Running my freelance missions with AI agents, I saw their limits first-hand: brilliant on a task, lost on a project. The solution felt within reach — so I built it. A hierarchy: a CEO that breaks down the mission, supervisors that coordinate, workers that produce. Each with a clear scope, rules, and accountability.

The choice that changes everything: 100% local-first. No SaaS, no server, no data leaving your machine. The app drives the user's own Claude Code session, and the entire project memory lives in versioned files — readable, auditable, reversible. In short: your project stays in your hands, right down to the tooling.

02What I'm building

Determinism wrapped around the unpredictable

  • Deterministic Rust layer — 19 crates: lock management, quotas, budget, processes, notifications, git mechanics. The AI proposes, the system enforces.
  • Filesystem vault — tickets, ADRs, reports, data contracts: everything is a file (JSON / JSONL / Markdown), validated by 16 JSON Schemas. No opaque database.
  • Claude Code driven as a subprocess — the app orchestrates the CLI directly, with real-time event aggregation and quota tracking.
  • Tauri desktop UI — React 19 + TypeScript in a WebView, generated Rust bindings, 21 specified screens.
  • Engineering discipline — 100+ ADRs (architecture decision records), versioned spec, cargo quality gates, Playwright tests.
03Stack
Rust19 crates — deterministic layer
TauriNative desktop app
React 19UI inside the WebView
TypeScriptBindings generated from Rust
Claude CodeAgent engine, CLI-driven
MCPVault server for agents
JSON Schema16 data contracts
PlaywrightE2E UI tests
04Why this project

My laboratory, your guarantee

Argus Forge is where I push what I believe in the furthest: documented, constrained, auditable systems — even with AI in the middle. Everything I learn here about agent orchestration, guardrails and local-first engineering flows straight back into my client missions. What I imagine, I code — and this time, the tool codes with me.