# Unpack > Open-source framework that turns AI research conversations into structured, buildable projects for Claude Code and Codex. Unpack is an open-source framework for AI-assisted development where conversations are the primary input. It bridges the gap between exploratory AI chat sessions and structured software development by transforming messy research conversations into phased build plans that coding agents can execute reliably. Research freely in ChatGPT, Claude, or any AI tool, then drop the conversation into your project. Unpack decompresses it into structured specs, phases, and execution plans that agents like Claude Code and Codex can follow systematically. - **Conversation-First Workflow** - Save any AI conversation as a markdown file and run /up-bootstrap to extract specs, requirements, architecture decisions, and create a phased build plan automatically - **Phase-Driven Execution** - Run /up-next to execute phases one at a time, each with clear scope, completion criteria, and tests so the coding agent follows the plan rather than improvising - **Snapshot and Iterate** - Export the current project state as a portable markdown file with /up-snapshot, take it to any AI for further research, then apply findings back with /up-apply - **Agent-Agnostic Design** - Works with Claude Code via CLAUDE.md and slash commands, and with Codex via AGENTS.md, using markdown instructions any agent can read - **Spec-Driven Documentation** - Maintains two documentation layers: agent docs in .unpack/docs/ as the source of truth for AI coding agents, and human docs in guide/ for developers and stakeholders - **Standards Library** - Ships with 15 coding standards covering TypeScript, React, Next.js, Express, Prisma, Tailwind, Python, Docker, and AWS, automatically matched to your stack during bootstrap - **Conversation Versioning** - Every conversation is preserved and numbered in .unpack/conversations/, making the full research history traceable so every decision can be traced back to its origin - **Long Conversation Handling** - Uses a two-pass triage protocol for complex conversations over 2000 lines, scanning for topic timelines, pivot points, and final decisions before extracting - **Existing Project Adoption** - Run /up-adopt to bring Unpack into an existing codebase, where the agent scans the project, generates discovery docs, and creates alignment phases - **Flexible Input Formats** - Accepts any conversation format including ChatGPT exports, raw AI conversation copy-paste, meeting notes, brainstorm docs, or design discussions in markdown To get started, use the GitHub template to create a new repo, clone it locally, and run /up-init to configure your agent and preferences. Research your project idea in any AI, save the chat as conversation.md, then run /up-bootstrap followed by /up-next to begin building. ## Features - Conversation-to-spec decomposition - Phase-driven build execution - Project state snapshots for external research - Claude Code and Codex agent support - Conversation versioning and traceability - Two-layer documentation system (agent docs and human docs) - 15 built-in coding standards for common stacks - Long conversation triage protocol - Existing project adoption workflow - Spec-driven development with Architecture Decision Records - Any-format conversation input support - ChatGPT conversation export bookmarklet - Mintlify-powered human documentation generation - Stack-aware standards matching - Portable markdown snapshots ## Integrations Claude Code, Codex, ChatGPT, Claude, Gemini, Mintlify, GitHub ## Platforms MACOS, LINUX, WINDOWS, DEVELOPER_SDK ## Pricing Open Source ## Links - Website: https://github.com/apresmoi/unpack - Repository: https://github.com/apresmoi/unpack - EveryDev.ai: https://www.everydev.ai/tools/unpack