A developer tool · v1 in trial
Stop letting AI assistants guess your codebase.
fewer tokens than a symbol-only indexer on a measured cleanup pass. Real per-team number lands at GA.
cix gives Claude Code, Codex, and Gemini a live, queryable index of your project — every function, route, table, and import edge — exposed as MCP tools they call before they generate. Source code stays on your machine. Only structural metadata is indexed.
01 — WHAT IT DOES
Search before writing. Schema before guessing.
cix gives your coding assistant a precise, up-to-date understanding of the project it's working in — exposed as a small set of MCP tools. The assistant queries the index before it generates, reads symbols instead of whole files, and stops inventing details that aren't there.
02 — TRUST
Source code never leaves your machine.
cix indexes the shape of your codebase, not its contents. Symbols, signatures, and call graphs cross the wire. The lines of code do not. Per-user index even inside team accounts — teams share billing and seats, never an index.
- ●source.code
- ●secrets
- ●git history
- ── stays here
- →symbols
- →signatures
- →call graphs
- ── metadata only
FIG. 01 — DATA BOUNDARY
03 — SPEND
Three places we save tokens.
01
Narrowed context
Claude reads only the symbols it needs, not whole files. Most prompts shrink by 60–80%.
02
Cache-friendly prompts
We structure context so Anthropic's prompt cache hits more often. Cache reads cost ~10× less than fresh tokens.
03
Fewer re-reads
Claude doesn't re-grep your repo on every turn. The index already knows where things live.
04 — DIFFERENCE
Measured on a real cleanup review.
Case study · 2026-04
Same codebase, same prompt, same evaluator. Two passes — one with a symbol-only indexer, one with cix. Laravel 12 API + Vue 3 SPA, ~300 indexed files, mid-rename.
fewer tokens than a symbol-only indexer
What cix surfaced that a symbol graph can't
- A hardcoded MySQL credential in a Python script left in /public — surfaced via get_project_orientation.
- Two stale duplicate-table migration warnings — emitted on every cix response, invisible to a symbol graph.
- Route parameter naming inconsistency between apiResource and custom routes — visible at a glance in list_routes.
PROOF
FIG. 02 — A FINDING THE BASELINE MISSED
“cix surfaced a hardcoded MySQL credential in a production-served public directory that a symbol-only indexer missed.”
Pre-release cleanup pass on a Laravel 12 + Vue 3 SPA. The credential lived in a forgotten Python WSGI script in /public — nothing imported it, so a symbol graph had no reason to surface it. cix flagged it because the orientation step scans for runtime database connections in route-served directories.
Read the cleanup case study →05 — COMPANION (BETA)
Scout — a local-model companion to cix.
Built for teams already running their own inference. Scout takes the high-volume, structured work — summaries, triage, audits — and runs it on your own hardware, so your frontier model is free for the changes that actually require its judgment.
cix is the core product. Scout extends it.
What it does
01
Routine work, on your own hardware
Change summaries, error triage, security sweeps, and convention audits run on infrastructure you control.
02
Grounded in the live index
Every answer is verified against cix before it appears, so suggestions don't drift into invented detail.
03
Keeps prompts focused
Hands smaller, structured prompts back to your primary assistant — work stays on cix's evidence rails.
Why it exists
Cost control
Most coding-assistant traffic is repetitive. Pushing that volume to a local model keeps frontier-model spend on the changes that need it.
Local-first by default
Routine analysis stays on hardware you control. Nothing about Scout's workload leaves the machine.
Faster iteration
Local inference responds quickly to small, structured prompts, so the loops that fire most often feel immediate.
When to use it
Reach for Scout if
- You already run Ollama (or a comparable local stack) and want it doing real work.
- You do enough code review, triage, and audit cycles that frontier-model spend is becoming a line item.
- You want a clean separation between high-judgment work and high-volume work.
Skip Scout if
- You're not running local inference and don't want to start.
- Your project is small enough that frontier-model spend isn't a concern yet.
- You'd rather keep your stack tight.
cix works fully without Scout. Scout is there for the teams that want it.
06 — PRICING
Pricing that pays for itself.
07 — INSTALL
One command.
macOS, Linux, Windows (WSL).
$ curl -fsSL https://usecix.com/install | sh