Your Agent
reads your code. calls tools. remembers. runs locally.

An agentic engine built in Rust. It plans multi-step actions, uses 13+ tools, and runs entirely on your machine with your API keys. No cloud dependency.

Fawx Fawx macOS App (Dark Mode) Fawx macOS App (Light Mode)

Built different.

A focused set of capabilities designed for real work, not demos.

01

Built in Rust

Starts in milliseconds. Runs for days without leaking memory. A single compiled binary with no interpreters, no runtimes, and no Docker.

$ fawx --version
fawx 1.0.0 (2026-03-19)
02

Persistent Memory

Fawx remembers across sessions. It learns your projects, preferences, and patterns using built-in embeddings for semantic search.

03

Native Tool Calling

13+ built-in tools for files, web, browser, git, and shell. The model calls them natively through provider APIs. No prompting tricks.

04

WASM Plugins

Hot-loadable WebAssembly skills with cryptographic signature verification. Sandboxed execution. Build your own in any language that compiles to WASM.

05

Kernel Safety

Compiled safety layer the AI cannot modify. Capability-based boundaries instead of permission prompts. The agent works freely within its space; hard blocks only on irreversible actions.

06

Native macOS & TUI

SwiftUI native app for macOS and iOS. Full terminal UI with ratatui. HTTP API for automation. Same engine, same conversation, any surface.

Fawx thinks,
then acts.

A real agentic loop. Fawx reads context, calls tools, and synthesizes answers. Autonomously, on your machine.

Works with Claude, GPT, or any OpenAI-compatible provider. Switch models mid-conversation. Token-by-token streaming shows you the thinking in real time.

fawx
Find the bug in the auth flow and fix it
🔧 read_file src/auth/oauth.rs
247 lines
🔧 search_text "callback" in src/auth/
3 matches
Found the issue. The callback URL validation on line
142 uses a prefix match instead of exact match,
allowing path traversal in the redirect URI.
🔧 write_file src/auth/oauth.rs
Fixed: exact match on callback path
🔧 run_command cargo test auth::
12 tests passed

Layered by design.

Three clean layers. The kernel enforces safety. The loadable layer provides intelligence. Shells are replaceable.

01
Shells
Native App · TUI · HTTP API
02
Loadable
Skills · Memory · Providers · Channels
03
Kernel
Loop · Gates · Tripwire · Ripcord

The agent can modify its own skills and behavior. The safety kernel that governs it is compiled into the binary and cannot be changed at runtime. Not by the user, not by the agent.

Download.
Setup. Go.

The setup wizard handles everything. You just need an API key from Claude or ChatGPT.

Download

Grab the DMG from the download page. Drag Fawx to Applications. Done.

Setup

Launch Fawx. The setup wizard walks you through connecting your AI provider: Claude, ChatGPT, or your own API key.

Chat

Start talking. Fawx handles tools, memory, and multi-step reasoning. It gets better the more you use it.

FAQ

Fawx works with Claude (Anthropic), ChatGPT (OpenAI), and any OpenAI-compatible provider. You can use your existing Claude or ChatGPT subscription, or bring your own API key. Switch models mid-conversation with a single command.
Fawx runs entirely on your machine. Your files, conversations, and memory stay local. API calls go directly from your computer to the AI provider you choose. Nothing routes through our servers. Credentials are encrypted at rest with AES-256-GCM.
ChatGPT and Claude are conversational interfaces. Fawx is an agentic engine that plans multi-step actions, calls tools on your machine (files, shell, git, web), remembers across sessions, and works autonomously. It uses those same models as its intelligence layer, wrapped in a real execution environment with safety boundaries.
Fawx is designed for developers and power users. The native macOS app provides a friendly GUI, but the tool system is built around code, files, and command execution. If you're comfortable in a terminal, you'll feel at home.
Fawx is free to download and use. You bring your own AI provider credentials, either through an existing Claude/ChatGPT subscription or your own API key. You pay the provider directly for model usage.