RUSVEL
The Solo Builder’s AI-Powered Virtual Agency One binary, one human, infinite leverage.
RUSVEL is an AI-powered virtual agency built with Rust + SvelteKit. It gives a single person the leverage of an entire agency through 14 autonomous departments, each powered by AI agents.
Quick Start
git clone https://github.com/mbaneshi/rusvel
cd rusvel
cargo run
# Open http://localhost:3000
What’s Inside
- God Agent — Your AI companion that knows your identity, products, and mission
- 14 Departments — Forge, Code, Content, Harvest, GTM, Finance, Product, Growth, Distribution, Legal, Support, Infra, Flow, Messaging
- Knowledge/RAG — fastembed + lancedb for semantic search over your documents
- Self-Improvement — The app can analyze and improve its own codebase
- 55 workspace members — Hexagonal architecture; 22 port traits in
rusvel-core/src/ports.rs(15 primary + 5*Store+ChannelPort+BrowserPort+RusvelBasePort) - ~645 tests (workspace sum); full
cargo testpasses in a normal dev environment
See Repository status for canonical metrics and links to docs/status/current-state.md on GitHub.
Architecture
God Agent (Chat — full authority + visibility)
├── Forge — Mission planning, goals, reviews
├── Code — Full Claude Code capabilities
├── Content — Draft, adapt, publish across platforms
├── Harvest — Find opportunities, score, propose
├── GTM — CRM, outreach, invoicing, deals
├── Finance — Ledger, runway, tax estimation
├── Product — Roadmap, pricing, feedback
├── Growth — Funnel, cohorts, KPIs
├── Distribution — SEO, marketplace, affiliates
├── Legal — Contracts, compliance, IP
├── Support — Tickets, knowledge base, NPS
├── Infra — Deploy, monitor, incidents
├── Flow — DAG workflow engine
└── Messaging — Notification channels
Each department is autonomous — own config, own chat, own agents, own events. God sees everything and can orchestrate any combination.
Stack
- Rust edition 2024, SQLite WAL, Axum, Clap 4, tokio
- SvelteKit 5, Tailwind CSS 4
- LLM: Claude CLI (Max subscription), Ollama, OpenAI, Claude API
- RAG: fastembed (local ONNX embeddings) + lancedb (vector search)