All member projects

Code · Built by members

LLMception

A prototype daemon that forks an AI coding agent at every ambiguous choice into parallel git worktrees, so you come back to answers instead of questions.

Conceptual illustration: A prototype daemon that forks an AI coding agent at every ambiguous choice into parallel git worktrees, so you come back to answers instead of questions.
AI-generated editorial illustration, not a product screenshot.

What it is

LLMception is a command line tool for long-running AI coding tasks. When you hand an agent a vague task and walk away, it usually either stops to ask a question or quietly picks one approach. LLMception takes a third path: at each decision point it explores every reasonable option at once.

Each option runs in its own git worktree and can branch again, building a decision tree whose leaves are complete implementations. When you return, you answer the questions, each answer prunes the branches you did not want, and you apply the one that is left. It is aimed at developers who run agents for long stretches and want to keep token costs in check.

What it does

  • Intercepts the questions Claude Code would ask and spawns a separate session for each answer option.
  • Keeps every branch isolated in its own git worktree.
  • Lets you walk the tree by answering questions, interactively or inline, then apply the winning implementation.
  • Caps the tree with limits on depth, width, node count, concurrency and spend.
  • Supports Claude Code on a subscription, the Anthropic API, the OpenAI API and local models through Ollama.
  • Shows exploration status and a full tree view, and cleans up worktrees afterwards.

Built with

  • TypeScript
  • Node.js
  • git worktrees
  • Claude Code
  • Anthropic API
  • OpenAI API
  • Ollama
  • agents
  • developer tools

More from the community

Code

governed-agent

by Jeremy

Capability-based security, scope-bounded planning and out-of-band audit for autonomous LLM agents.

App

ArchLang

by Chan

An open-source language for floor plans, now also inside ChatGPT: ask for a plan in plain words and ChatGPT writes it in ArchLang, compiles it and checks it.

App

Remote Clipboard

by Ming

Copy on one device and paste on another, or share with people and agents, through a short-lived room or the rclip CLI.

Back to all member projects