Stop letting AI assistants guess your codebase

AI coding assistants are good at writing code. They are bad at remembering yours.

They duplicate helpers that already exist. They drop files in the wrong folders. They invent column names. They miss the dead code, the broken references, the forgotten credentials sitting in production-served directories.

cix gives them a real map of your project so they stop guessing.

It builds a live, queryable index of your codebase — every function, class, route, table, and import edge — and exposes it to your AI assistant so the assistant searches before it writes, reads only what it needs, and stays within the structural rules your team has agreed on.

Measured impact

In a side-by-side cleanup pass on a real Laravel + Vue project:

About half the tool calls (29 vs. ~55).
About a third of the tokens (30–40k vs. 80–100k).
Three additional structural issues caught that the indexer-free pass missed — including a hardcoded MySQL root credential in a Laravel public/ directory.

The cleanup workflow is the strongest case for cix. Concrete, scary, measurable. Read the case study →

Three promises

1. The assistant searches before it builds. When asked to add a feature, it asks the index "does something like this already exist?" instead of generating a duplicate. Existing helpers, components, and utilities surface in one query.

2. The assistant respects your structure. File placement, naming conventions, folder rules — encoded as machine-readable conventions and checked before files are written. On Claude Code, where write hooks are most mature, violations are blocked at the file system level. On other clients, violations surface as warnings the assistant has to address before continuing. Either way, the rule lives in the project, not in someone's head.

3. The assistant tells you what it doesn't know. Every answer carries a confidence signal. When the index is stale, when a file failed to parse, when a query couldn't be resolved — the assistant says so instead of guessing.

What you actually get

One command per machine, one command per project, five minutes total.
Works with Claude Code, Codex, and Gemini CLI. Same index, same conventions, all three clients.
Convention profiles for ten common stacks ship out of the box: React, Vue, Next.js, Django, Flask, FastAPI, Laravel, Rails, Express, plus a generic profile.
A schema view built directly from your migrations. The assistant stops inventing column names.
A change-impact view that answers "what breaks if I rename this?" in one call instead of ten greps.
Honest behavior when something is unsupported. If your project is in a language without a parser, the system says so — it does not fabricate results.

Start here based on your role

For evaluators — technical decision-makers running a real assessment. Includes a one-week evaluation plan, what to measure, and the per-project benchmark data behind our claims.
For engineering teams — managers, leads, and buyers thinking about adoption. The team-scale problem, the risk it reduces, the cost of doing nothing, and what success looks like.
For solo developers — individual users on personal projects, side work, or single-maintainer codebases. The "I'll remember" trap, what it fixes for you, and where it isn't worth the setup.

Or dive deeper

New here? Read Overview for a longer pitch, then Why cix for the problem this solves.
Want the technical shape? How it works explains the architecture without the implementation noise.
Checking your stack? Language support — full / partial / unsupported tiers, no guessing.
Curious about the trade-offs? Limitations is the most important page on this site.

Plain-spoken summary

cix is an index. An AI assistant plus a real index of your code is materially better at working in your project than the same assistant alone. That is the whole pitch. The rest of these pages explain why that matters and what it looks like in practice.