Why Fawx exists.

AI assistants today run on someone else's server, through someone else's interface, with someone else's rules. Your conversations live on their infrastructure. Your data passes through their systems.

Fawx runs on your machine, with your API keys, under your control. An agentic engine that plans, acts, and remembers.

What makes Fawx different

• Local-first. Your data never leaves your machine unless you explicitly tell it to. Conversations, memory, and credentials are all local, all encrypted.
• Agent, not assistant. Fawx doesn't just respond to prompts. It decomposes tasks, calls tools on your machine, reads and writes files, runs commands, and synthesizes results across multiple steps.
• Kernel safety. A compiled safety layer the AI cannot modify, inspect, or bypass. Real enforcement compiled into the binary, not prompt-based guardrails. The agent works freely within defined boundaries; the kernel ensures those boundaries hold.
• Self-extending. WASM skill plugins let Fawx learn new capabilities without rebuilding. Skills are cryptographically signed, sandboxed, and hot-reloadable.
• Multi-surface. The same engine powers the native macOS app, the terminal UI, and the HTTP API. Start a conversation in the app, continue it in the terminal. Same thread, same memory.

Where it's going

The architecture is designed for what comes next: distributing work across multiple machines, running local models with zero cloud dependency, and eventually, an operating system where the AI is the environment you work in.

iOS is in testing. Android is planned. The engine is the constant; the shells change.

The philosophy

Every AI tool asks you to trust it. Fawx earns trust by running on your hardware, keeping your data local, and enforcing safety through code.

The kernel is the foundation. Powerful tools need real constraints, and those constraints need to be verifiable.