# codemix

> codemix is an AI product intelligence platform that captures product intent into a semantic graph, then steers coding agents like Claude Code, Cursor, and Copilot through planning, implementation, and code review.

codemix is an AI product intelligence platform that gives teams and AI coding agents a shared, always-current source of truth for how their product is supposed to behave. It captures intent — business domain concepts, user flows, rules, edge cases, and constraints — into a semantic graph that sits between product decisions and code, then carries that context through planning, implementation, and review so agents build what was actually meant rather than what they imagined.

Teams start by either running a guided requirements interview for a new product or importing an existing git repository. codemix analyzes the codebase to build a semantic map of concepts, actions, events, and screens, then keeps that map synchronized as the product evolves. Changes are proposed and refined the way teams naturally work: through chat, visual diagrams of architecture and user flows, and collaborative editing of proposals that behave like pull requests for intent rather than code.

Once a proposal is approved, codemix generates concrete implementation tasks with acceptance criteria, which can be sent to Linear or Jira, handed to connected coding agents, or both. During execution, codemix answers agent questions using the product graph, reviews changes against intent before they land, and catches drift between what was built and what was meant. When implementation reveals new edge cases, codemix proposes spec updates so the source of truth stays current.

- **Living product spec**: *A semantic graph of the product's domain model — concepts, actions, events, and screens — kept synchronized with the codebase automatically.*
- **Guided requirements interview**: *Turns product ideas into a detailed PRD reflecting what you actually want to build before writing any code.*
- **Codebase import and analysis**: *Connects a git repository and reverse-engineers how the product actually works today into the product graph.*
- **Chat and visual canvas**: *Describe changes in plain language or draw architecture, user flows, and wireframes on a shared canvas that drives real product updates.*
- **Proposal-based intent changes**: *Pull-request-style workflow for intent: scope a change, review it, tighten it, then turn it into implementation tasks.*
- **Agent steering**: *Pushes tasks to Cursor, Claude Code, GitHub Copilot, Codex, Gemini, Lovable, v0, Bolt, Windsurf, and Replit with the context they need to build the right thing.*
- **Task routing to Linear and Jira**: *Generated implementation tasks flow into existing trackers so teams keep their current workflow.*
- **Code review against intent**: *Reviews proposals and code changes against the product graph before merge, flagging where built behavior diverges from intended behavior.*
- **CLI for terminal and automation**: *Install via `npm install -g codemix` for project-scoped Q&A, local discovery, and scriptable task workflows in CI.*
- **Remote MCP server**: *OAuth-scoped access lets any MCP-compatible agent query the product graph directly during development.*
- **Slack bot**: *Answers product-aware questions in Slack channels with references back to source specs.*
- **Open source graph engine**: *The underlying @codemix/graph — a type-safe, realtime, Cypher-queryable graph database inside a Yjs CRDT — is available as an MIT-licensed npm package.*

## Features
- Living product spec kept current automatically
- Semantic product graph for humans and AI
- Guided requirements interview for new products
- Existing codebase import and analysis
- Collaborative chat for proposing product changes
- Visual canvas for architecture, user flows, and wireframes
- Pull-request-style intent proposals with review
- Automated implementation task generation
- Agent steering during development with grounded answers
- Code review against product intent before merge
- Automatic spec updates when implementation reveals edge cases
- CLI for project-scoped Q&A and task automation
- Remote MCP server for agent access
- Slack bot for in-channel product answers
- Open source @codemix/graph engine (MIT)

## Integrations
Cursor, Visual Studio Code, Claude Code, GitHub Copilot, OpenAI Codex, OpenCode, Windsurf, Replit, Lovable, Bolt, v0, Linear, Jira, Notion, Confluence, Slack, GitHub

## Platforms
WEB, CLI, API

## Pricing
Freemium — Free tier available with paid upgrades

## Links
- Website: https://codemix.com
- Documentation: https://codemix.com/docs
- EveryDev.ai: https://www.everydev.ai/tools/codemix