Features
cix's surface is intentionally narrow. Each feature exists because, without it, the AI assistant would have to either guess or read its way to an answer. Together they replace a large chunk of the manual exploration work that happens silently in every coding session today.
This page is the index. Each feature has a deep-dive page linked below.
Find before you build
The single most common failure mode of AI assistants — duplicating helpers, components, and utilities that already exist — is also the easiest to prevent. cix gives the assistant a fast, structured way to ask "does something like this already exist?" before generating a new one.
The query understands word boundaries (a search for date returns formatDate and dateUtils, not update or validate), respects case styles, and returns precise file/line locations.
Convention enforcement
You decide where files belong and how they should be named. cix encodes those decisions as machine-readable rules and enforces them at the moment of file creation.
On clients that support write hooks, violations are blocked outright — the assistant cannot put a controller in the components folder, because the file system never accepts the write. On other clients, the rules still run and surface as warnings the assistant must address.
Ships with default rules for ten common stacks; tunable per project.
Schema awareness
When an assistant writes a database query without seeing the schema, it invents column names. When it writes a model without seeing existing models, it duplicates fields under different names.
cix parses your migrations — Laravel, Django, Alembic, SQLAlchemy, Prisma, raw SQL — and exposes the resulting schema as a single query. Every model, query, and migration the assistant writes can ground itself in the actual table structure rather than a guess.
Impact analysis
Before changing a function, route, or table, the assistant can ask: "what depends on this?" The answer comes back as a structured report — direct callers, transitive callers, affected tests, affected request surfaces, and a risk tier with reasons.
This is the capability that turns "let me try this rename and see what breaks" into "here are the seventeen places this rename touches; let me handle them deliberately."
Project orientation
The first turn of any session is wasted if the assistant doesn't know the basics — what framework you're using, where the entry points live, what the major directories are. A single orientation query returns all of that in one structured answer.
For new projects (or new contributors), this is a five-minute onboarding compressed into a single tool call.
Multi-assistant support
cix is built on the Model Context Protocol, the open standard several AI coding tools have adopted. One install configures Claude Code, Codex, and Gemini CLI to share the same index and the same conventions.
Switching assistants — or running a team where different people use different ones — does not require maintaining separate context for each.
Honest behavior
Every answer the system returns includes a confidence signal. When the index is stale, when a file failed to parse, when a query couldn't be resolved against a supported language, the system says so explicitly.
This sounds like a small thing. It is not. The alternative — a system that returns a confident-looking answer when it doesn't actually know — is how AI tools earn their reputation for fabricating details.
cix would rather say "I don't know" than make something up. That choice is baked into the design.
What's underneath all of this
A live, queryable index of your codebase that updates as you work. The features above are different ways of asking that index questions. The index handles the storage; the queries handle the questions; the conventions handle the rules; the assistant uses all three through one integration.
For the architectural picture, see How it works.
See it in motion
- Workflows — common end-to-end usage patterns.
- Pre-release cleanup case study — measured side-by-side with a less-capable indexer.
- 25-repo benchmark — how the features hold up across diverse real codebases.