Sorcerer icon

Sorcerer

Sorcerer is a local-first macOS app for simulating how AI-driven character communities may react to a what-if scenario. It helps users explore hypotheses with contextual briefs, configurable communities, branching simulations, polls, and cited reports.

Sorcerer

What Sorcerer Does

Sorcerer is a local-first SwiftUI application for macOS that simulates a community of AI-driven characters around a hypothetical question. Users define a scenario, provide relevant context, assemble a community, and observe how posts, reactions, opinions, and actions develop across rounds.

The tool is intended for exploring possible reactions, objections, and consequences rather than establishing real-world evidence. It can use Claude through the Claude CLI, ChatGPT through the Codex CLI, or a locally running Ollama instance, while retaining experiment data as local JSON files.

Simulation tools

Question-led experiments

Start with a what-if question, select a setting such as a social feed, forum, group chat, town square, or workplace, and configure rounds and pace. The app provides a live preview while the experiment is being set up.

Context briefs with knowledge boundaries

Build a brief from a GitHub repository through your gh login, a local folder, a website, or files. Facts are labeled by who knows them, including public, insider, and hidden knowledge. The stated reader excludes .env files, keys, credentials, node_modules, and build output.

Persistent character communities

Create or reuse communities and edit members before a run. Character profiles include personality, interests, goals, relationships, and memories; characters can reply, react, and weight attention toward people they are tied to.

Live rounds, actions, and views

Watch posts arrive round by round in a feed, switch to a relationship map or mood graph, and track listed actions such as choices supplied by a template or proposed from the question. Participants can also leave, return, or invite others into the world.

Polls, branching, and cited reports

Ask the community a poll with custom options, either off the record or as a public event that affects the next round. Fork at a chosen round to explore another path, then use a report that compares branches and links findings to source posts.

Practical uses

  • Explore a product-change hypothesis

    Model how a fictional customer community could discuss a proposed subscription increase, including possible upgrades, cancellations, and poll responses. Treat the results as prompts for research rather than demand evidence.

  • Test communications scenarios

    Give the community a headline, announcement, or event and observe how responses can circulate through relationships in a group-chat, forum, or social-feed setting. Inject follow-up events to test how the discussion changes.

  • Reason from controlled context

    Load documentation from a repository, local folder, website, or files into a brief, distinguish common from insider or hidden information, and examine how each knowledge boundary changes a simulated discussion.

  • Compare alternate outcomes

    Fork an experiment at a previous round and alter a decision or event while retaining the community and its memories up to that point. Compare the branches through the report and its cited posts.

Pros and Cons

Pros

  • Uses existing Claude, ChatGPT/Codex, or Ollama access rather than bundling a separate model service.
  • Keeps experiments, reusable contexts, and settings locally in readable JSON files.
  • Supports alternate branches from an earlier round, with reports that compare divergence and cite underlying posts.
  • Lets users control what a community knows through public, insider, and hidden context labels.

Cons

  • Simulations are framed as hypotheses to investigate and do not replace evidence from the real world.
  • The native app requires macOS 15 or later.
  • Using Claude or ChatGPT depends on the corresponding CLI being installed and signed in; provider usage limits can pause runs.

FAQ

What platform does Sorcerer support?

Sorcerer is a native macOS application. The latest release states it supports macOS 15 or later and is distributed for both Apple silicon and Intel Macs.

Which AI providers can Sorcerer use?

It uses AI tools you are already signed in to: the Claude CLI (Claude Code), the Codex CLI for ChatGPT, or Ollama running locally at localhost:11434. You can select models by provider and assign different models to characters, reports, and one-to-one chats.

Where are simulations stored and can they be exported?

Sorcerer stores its data in ~/Sorcerer as plain JSON, including experiments, reusable context briefs, and settings. It can export an experiment with its full history or export a fresh scenario without history for another person to run.

What is the basic simulation workflow?

Create an experiment by defining a question and rules, providing context from supported sources or files, and setting up a community. Rounds then stream into the feed; you can pause, inject events, run polls, and fork a branch from a prior round.

Quick Facts

Category
AI simulation tool
Platform
Native macOS app; latest release lists macOS 15 or later
AI connections
Claude CLI, ChatGPT/Codex CLI, or local Ollama
Data storage
Local plain JSON in ~/Sorcerer
License
MIT
Source
github.com/gbrasil720/sorcerer