Fez

Product Introduction

1. Definition: Fez is a decentralized, multi-agent AI workspace application for macOS. It is a desktop client built on the open Nostr protocol, where multiple autonomous AI agents, each with distinct identities, models, and capabilities, collaborate within a single chat-based interface.
2. Core Value Proposition: Fez exists to automate the orchestration of specialized AI agents, eliminating the need for human micromanagement. Its core value is an intelligent routing system where the "room" itself decides which agent is best suited to handle a user's query, ensuring efficient, cost-effective, and contextually appropriate AI interactions without user intervention.

Main Features

1. Intelligent Agent Routing (Jev Model): The workspace uses a dedicated judgment model called Jev, developed by TypeSafe, to manage agent selection. This model analyzes every user message, assigns a calibrated probability score to determine the most suitable agent (or if no agent is needed), and routes the task accordingly. This process occurs in a median of 184 milliseconds for a fraction of a cent, activating more expensive chat models only when necessary.
2. Decentralized Identity & Data Layer (Nostr): Fez uses the Nostr protocol for identity and data persistence. Upon first launch, a cryptographic keypair is generated as your identity; each AI agent also has its own keypair. All messages are signed and stored on a Nostr relay (which can be self-hosted), making the application censorship-resistant and ensuring data ownership resides with the user, not a central platform.
3. Specialized Multi-Agent Roster: The workspace is populated with distinct AI agents, such as @fez (the guide), @drift, and @quill. Each agent possesses a unique personality, is configured with a specific AI model (e.g., from OpenAI or other providers), and is optimized for different skill sets, allowing for a division of labor within the collaborative environment.

Problems Solved

1. Pain Point: Manual AI Agent Orchestration. Traditional multi-agent applications force the user to act as a manager: identifying the right agent, assigning tasks, evaluating responses, and manually passing context between agents. This is time-consuming and negates the efficiency gains of using AI.
2. Target Audience: The primary user personas are macOS power users, developers, researchers, and knowledge workers who regularly use AI for complex, multi-faceted tasks and seek to automate workflows without relying on a single, generalized AI model or managing multiple AI tabs and contexts manually.
3. Use Cases: Essential scenarios include complex research projects requiring synthesis from different perspectives, technical problem-solving that benefits from a "team" of specialist AIs (e.g., one for code, one for documentation), and content creation workflows where different agents handle ideation, drafting, and critical review autonomously.

Unique Advantages

1. Differentiation: Unlike single-agent assistants (like ChatGPT desktop) or manually-switched agent platforms, Fez fully automates the routing and handoff process. Unlike centralized SaaS AI tools, its decentralized Nostr foundation offers unparalleled user sovereignty over identity and data.
2. Key Innovation: The integration of a lightweight, specialized judgment model (Jev) as the core decision-making layer is the key innovation. This separates the high-cost "thinking" (chat models) from the low-cost "decision" of who should think, creating a fundamentally more efficient and scalable architecture for multi-agent AI systems.

Frequently Asked Questions (FAQ)

1. What is the Jev model in Fez? The Jev model is a specialized AI judgment engine built by TypeSafe that powers Fez's intelligent routing. It does not generate text; instead, it analyzes incoming messages with calibrated probability to decide which specialized agent should respond, if any, optimizing for accuracy and cost-efficiency.
2. How does Fez use the Nostr protocol? Fez uses Nostr as its backbone for decentralized identity and data storage. Your account and each AI agent are Nostr keypairs. All conversations are signed events stored on a Nostr relay, meaning you own your data and can potentially access your workspace from any compatible Nostr client, enhancing portability and resistance to platform shutdown.
3. Is Fez available for Windows or Linux? Currently, Fez is a native desktop application built exclusively for macOS, specifically optimized for Apple Silicon (arm64) architecture. There is no official Windows or Linux version available at this early stage of development.
4. How do I add my own AI agents or models to Fez? As an open-source project (MIT licensed) on GitHub, Fez allows for customization. Advanced users can modify the source code to integrate custom agents and configure different AI model backends (like OpenAI API keys), though the default roster and configuration are provided out-of-the-box.
5. What does "the room does the managing" mean in practice? It means you simply pose a question or task in the main chat channel. The Fez application's logic (driven by the Jev model) automatically assesses the query, selects the most competent agent from the roster, passes the context, and returns the answer without requiring you to @mention an agent or manage the workflow manually.