Sorcerer

Product Introduction

1. Definition: Sorcerer is a native, local-first macOS application (built with SwiftUI) that functions as an AI agent simulation sandbox. It is technically categorized as a desktop-based multi-agent simulation platform for hypothesis testing and scenario exploration.
2. Core Value Proposition: Sorcerer exists to enable users to cheaply, privately, and interactively simulate the potential reactions of a community to a "what if...?" question. Its core value is providing a structured, visual environment to model complex social dynamics, opinion formation, and consequence exploration using the AI subscriptions (Claude, ChatGPT) or local models (Ollama) a user already possesses, without requiring API keys or a centralized service.

Main Features

1. AI Agent Community Simulation: Sorcerer creates a persistent world of AI-driven characters with defined personalities, goals, relationships, and memory. The simulation engine runs in discrete rounds within a configurable environment (e.g., social feed, forum). Agents autonomously post, react, remember past interactions, and are influenced by their social ties, creating emergent narrative dynamics. The technical implementation involves structured prompt engineering to maintain character consistency and context-aware memory recall across rounds.
2. Local-First, Bring-Your-Own-AI Architecture: All experiment data—including character profiles, conversation history, and world state—is stored as plain JSON files in ~/Sorcerer on the user's Mac. The app interfaces directly with local command-line tools: the claude CLI (for Anthropic's Claude models), the codex CLI (for OpenAI's ChatGPT models), or a local Ollama server. This architecture ensures data privacy, eliminates ongoing costs for the user, and leverages existing AI subscriptions.
3. Interactive Experiment Control & Forking: Users are not passive observers. They can pause simulations, inject custom events, conduct off-the-record or public polls (⌘P), and have 1:1 chats with any character. A critical feature is timeline forking (⇧⌘F), allowing users to create a divergent branch from any round to test alternative outcomes. The app provides a visual replay mode and comparative reporting between branches, with every insight directly cited to source posts.

Problems Solved

1. Pain Point: Traditional brainstorming and impact analysis are often limited by groupthink or inability to model complex, multi-stakeholder reactions over time. Product managers, writers, and strategists lack tools to stress-test ideas in a dynamic, conversational context before real-world deployment.
2. Target Audience: Primary personas include Product Managers & Founders testing feature reception, Content Creators & Writers exploring narrative outcomes, Community Managers modeling discussion trends, and Strategic Planners conducting lightweight scenario analysis. It also appeals to AI enthusiasts interested in multi-agent systems and emergent behavior.
3. Use Cases: Essential for simulating user community reaction to a pricing change, gauging internal team sentiment on a policy shift, brainstorming plot developments for a story, stress-testing a new feature announcement, or exploring the societal impact of a hypothetical event in a controlled, repeatable sandbox.

Unique Advantages

1. Strengths & Limitations (Pros & Cons):

   • Pros: Truly private and local-first; no data sent to a third-party server unless using Claude/ChatGPT CLI for processing. Zero marginal cost for users with existing AI subscriptions. Offers deep interactivity (forking, events, polls) not found in simple chat interfaces. Native macOS app provides a smooth, integrated user experience with system shortcuts.
   • Cons: Exclusively available for macOS 15+, limiting potential user base. Requires technical comfort to install and configure local AI CLIs. Simulation fidelity and character consistency are inherently dependent on the underlying AI model's capabilities and the user's prompt engineering within the app's framework. Not designed for quantitative forecasting.

2. Key Alternatives & Differentiation:

   • MiroFish: A direct inspiration, but MiroFish is typically a web-based service. Sorcerer differentiates by being a native desktop application with a local-first data model, offering greater privacy and integration. Sorcerer also emphasizes user interactivity and forking within a simulation.
   • General-Purpose AI Chatbots (ChatGPT, Claude): While these can role-play, they lack Sorcerer's structured multi-agent environment, persistent memory per character, visual timeline, comparative branching, and dedicated tools for community-wide polls and event injection. Sorcerer provides a framework for sustained, multi-participant simulation.
   • Agent Frameworks (AutoGen, CrewAI): These are developer-focused Python frameworks for building custom AI agent workflows. Sorcerer is a consumer-facing end-user product with a polished GUI, pre-built simulation mechanics, and no coding required, trading customizability for accessibility and a curated experience.

Frequently Asked Questions (FAQ)

1. Is Sorcerer AI free to use? Yes, the Sorcerer macOS application itself is free and open-source (MIT license). It operates on a "bring your own AI" model, meaning you must provide the AI processing power via your own subscriptions to Claude (via Claude Code) or ChatGPT (via Codex CLI), or by running free, local models through Ollama. There are no fees to the Sorcerer developers.
2. How does Sorcerer app ensure data privacy? Sorcerer employs a local-first architecture. All your experiments, character data, and conversations are stored as plain JSON files in a folder on your Mac (~/Sorcerer). When using Ollama, all processing happens locally. When using Claude or ChatGPT CLIs, only the necessary prompt data for a given simulation round is sent to the respective provider's API via your authenticated local client.
3. What AI models are compatible with Sorcerer? Sorcerer is compatible with models accessible through its supported providers: Claude models (Haiku, Sonnet, Opus) via the Claude CLI; any model available through your ChatGPT plan via the Codex CLI; and any model you have installed locally via an Ollama server (e.g., Llama 3, Mistral, Gemma).
4. Can Sorcerer predict real-world outcomes accurately? No. Sorcerer is not a forecasting tool. It is a simulation sandbox designed to surface a range of possible reactions, perspectives, and emergent narratives based on the initial conditions you set. Its outputs are hypothetical explorations meant to inspire thought, identify potential objections, and model dynamics, not to provide guaranteed predictions.
5. What are the system requirements for Sorcerer? Sorcerer requires macOS 15 (Sequoia) or later, and either an Apple Silicon or Intel Mac. You must also have at least one of the following installed and configured: the official claude CLI (Claude Code), the codex CLI, or a local Ollama server.