codemix is a semantic layer for software products.
It reads the product hiding inside a codebase: concepts, commands, screens, flows, events and the relationships between them. That model becomes something you can browse, query and use when planning real changes.
The point is simple: stop rebuilding intent from tickets, code archaeology and half-remembered Slack threads every time somebody asks "what should this do?"
Two useful paths
Use codemix when you want shared product understanding:
- create a project
- connect a repository or start from scratch
- run discovery
- inspect the domain map
- ask project-specific questions
- turn a change request into a proposal and task plan
Build with codemix when another tool needs that context:
- use the CLI for terminal workflows and task execution
- use the remote MCP server when AI clients need OAuth-scoped project context
- use API keys for automation that runs outside the browser
- use Slack when the team asks questions in channels
What the product model contains
Discovery looks for the nouns and behaviours that matter to a product, not just the files that implement them.
The important pieces are:
- concepts: domain nouns such as Organization, Project, Proposal, Task or Credit
- commands: actions that change state or perform side effects
- events: facts that happened and matter later
- screens: user-facing surfaces and what they display
- flows: end-to-end behaviours across screens, commands and concepts
This is why codemix answers product questions better after discovery. It has a model to reason from instead of a pile of disconnected snippets.
The main workflow
The normal loop is:
- discover the current product
- ask questions until the intent is clear
- describe a change
- review the proposal and affected areas
- generate tasks
- claim and execute work
- review the result against the plan
codemix is not a hard gate. It is a shared reference point. Project phases guide the team, but they do not replace judgement.
Where each surface fits
Use the web app for setup, browsing, chat, proposals, review flows, task planning and organization settings.
Use the CLI when you are already in a terminal, want project-scoped Q&A, need local discovery, or want a clean task execution loop.
Use Slack when questions happen in team channels and should keep thread context.
Use the MCP server when an AI client should ask codemix directly instead of guessing from its own prompt context.
Use the API docs when you are choosing credentials, automation shape or integration boundaries.
Read next
Start with Getting started if this is your first project.
Read API if you are wiring codemix into another tool.
Read CLI if you want terminal workflows.
Read MCP server if an AI client needs project context.
Read Slack if the team will ask questions from channels.
Conventions
Use lowercase codemix.
Use "Integrations" in product UI unless the MCP-specific detail matters.
Treat credits as the unit for AI-heavy work. Some plan limits affect what a team can do, especially hosted discovery, collaboration, Slack and code review on Solo.