Browse docs

← All docs

docs / features.md

Features

cix gives your AI assistant a set of tools for reading your codebase by structure rather than by guesswork. This section covers what those tools do and when each one earns its place.

You don't call these yourself. Your assistant picks them while it works — the value of knowing them is recognising when it should have reached for one, and asking for it by name.

The tool groups

GroupAnswersPages
OrientationWhat is this project, and what changed?Project shape, routes, schema, health
Search and lookupWhere does this live, and what does it say?Symbols, text, files, line ranges
Tracing and impactWhat else does this touch?Callers, data flow, blast radius
Validation and writingIs this right, and can it be changed safely?Conventions, audits, structural edits
CollaboratorsIs this repo already covered by someone else?Reading an owner's index
Branches and worktreesDoes this answer match the branch I'm on?Per-branch index, worktrees, uncommitted work

Two of these are less common elsewhere and worth knowing about: cix is anchored to git, so answers are scoped to the branch you're on and worktrees stay separate — and a repository someone else already indexed may need no work from you at all.

The other half: workflows

The groups above are what your assistant can look up. They answer questions about your code, one question at a time.

Workflows are the other half — how a piece of work gets run. A perfect index does not stop a long agent session from drifting off the task, or from declaring something finished without evidence. Those are failures of process, not of lookup, and no amount of retrieval fixes them.

WorkflowAnswers
The loopHow does this task get executed without drifting or ending in an unverified "done"?
ResearchWhat did we establish, and how do we know?
DesignWhat should be built, and what has to stay true?
PlansIn what order, and what is actually finished?
TidyWhat state is this repository really in?

Research, designs, and plans are stored with your cix account and are part of a paid plan. See Workflows for how they fit together and what each one needs.

Why structure beats reading files

An assistant without an index answers by opening files it guesses are relevant. That works until the codebase outgrows what fits in a context window, at which point the guesses quietly get worse — and a confident answer built on three of the eight relevant files reads exactly like a correct one.

An index changes the question from "which files should I open" to "where is this symbol defined, who calls it, and what breaks if it changes." Those have exact answers.

What this means in practice

  • Before writing something new, your assistant can check whether it already exists rather than adding a second implementation.
  • Before changing something, it can list the callers instead of hoping there aren't any.
  • Before creating a file, it can check where that kind of file belongs in your project.
  • After time away, it can summarise what changed structurally rather than replaying the commit log.

What your plan includes

Most of this works on the free tier: searching symbols, reading a symbol or a line range, listing files, reading your schema, checking conventions, finding callers, and the structural edit tools.

Twelve capabilities are grouped onto the paid plans, and pages below mark them where they appear:

PlanWhat it addsCapabilities
IndividualChange code safelyImpact analysis, counterfactual, file dependencies
ProStay up to dateCatch-up briefing, vital sign, pre-commit impact, last diff
MaxAnalyze and transform architectureConvention audit, stale symbols, rename detection, data flow, bulk rename

Each plan includes everything below it. See pricing for the full comparison.

Note — your assistant is only offered the capabilities your plan covers, so a gated tool is absent rather than failing when called.

Next: Orientation and project shape.