Your AI Reads
Your Architecture
Stop feeding entire files. Start indexing structure. Codebase Mirror is a cognitive implant - it scans your project, extracts the structural graph, and gives your AI a 30-token map instead of 50 files. Built into QuoxCORE's reflection system.
In plain words
What it is, where it lives, when to reach for it
- What is it
- A dashboard feature of QuoxCORE that scans a project and builds a compact structure map for agents.
- Where do I use it
- In the dashboard, on the Codebase Mirror page; agents read the map through the reflection system.
- When would I use it
- When agents keep reading whole files to answer questions your project structure already answers.
- How do I use it
- Open Codebase Mirror in the dashboard, enter a project path, and click Scan.
QuoxCORE is the free, self-hosted platform underneath this. What is QuoxCORE
A cognitive implant, not just a tool
Codebase Mirror is wired into QuoxCORE's reflection system - it's how your AI thinks about code.
Structure as knowledge, not files as data
Traditional code context dumps entire files into the AI's context window. Codebase Mirror takes a different approach: it pre-computes the structural graph - components, routes, registries, exports, dependency chains - and injects a compressed representation (~30 tokens) into every prompt.
This is the cognitive implant pattern. The AI doesn't read your codebase - it already knows the shape. When it needs detail, it queries the index for specific answers. The result: faster, more accurate responses with dramatically less token waste.
- Pre-computed structural graph injected into every prompt
- ~30 tokens replaces reading 50+ files
- Query API for on-demand detail: scan, index and diff are the mature surface, structured query is early
- Drift detection keeps the implant current
- Designed to reach every QuoxCORE agent from one install, rather than per agent
Without Mirror
AI reads raw files one by one, guessing which ones matter
~50 filesread per session to understand the projectWith Cognitive Implant
Structural awareness injected automatically - agents ask targeted queries
~30 tokensof structural context per promptScan. Index. Query.
Three steps from messy codebase to instant structural awareness
Scan
Auto-detects project type (React, Go, TypeScript, Rust, Python). Extracts component registries, route tables, exports, and dependency graphs.
Index
Stores structural data in SQLite. Computes diffs against previous scans. Tracks drift - new files, deleted modules, changed exports.
Query
Agents ask on demand: "where is the auth middleware?" or "list all API routes." ~30 tokens injected as structural context per prompt.
POST /api/v1/codebase/scan { "path": "quox-website" }
{
"project": { "id": "b9e75d2c4bee847c", "name": "quox-website", "type": "react" },
"metrics": {
"fileCount": 475,
"moduleCount": 48,
"routeCount": 450,
"serviceCount": 5,
"testFileCount": 0,
"registryCount": 1
}
}
invariant checks
FAIL plugin-manifest-sync Missing manifests: mikrotik-routeros
# That failure is the point. A scanner that only ever reports success
# proves nothing - this one caught a real drift the same night it ran.
GET /api/v1/codebase/b9e75d2c4bee847c/query?q=where is the auth middleware
{ "results": [] }
# Structured query coverage is thin today: this real question came back
# empty. Scan, index and diff are the mature surface; query is early.
Structure-first intelligence
Everything your AI needs to understand your codebase without reading it
Auto-Detect Project Types
Identifies React, Go, TypeScript, Rust, Python, and more. Extracts registries, routes, exports, and dependency graphs automatically.
Drift Alerts
Detects when the index is stale. New files, deleted modules, changed exports - flagged before your AI reads outdated context.
Cognitive Implant
Injects ~30 tokens of structural context into every AI prompt. The model sees your architecture without reading a single file.
Scheduled Scans
Run on commit hooks, cron, or CI. Index updates happen in the background - always fresh, never blocking.
Remote Fleet Scanning
Scan codebases on remote hosts via QuoxBastion. Index production deployments without cloning repos locally.
CODEBASE_MAP.md Generation
Exports the full structural index as a markdown file. Readable by humans, parseable by any LLM or MCP tool.
Layered by design
Each layer does one job and talks only to its neighbors
Feature Page
This page - explains the cognitive implant concept
Dashboard
Visual index browser, drift alerts, scan history
Context Injector
Injects structural tokens into AI prompts (~30 tokens)
Query API
Agents ask specific questions - "where is the auth middleware?"
Storage (SQLite)
Indexed structure, diff history, project metadata
Scanner Plugin
Detects project type, extracts registries, builds graph
Part of the QuoxCORE platform
Works with everything you already use
- QuoxCORE dashboard - visual index browser
- QuoxMCP - tools query the index via MCP protocol
- QuoxAgent - fleet-wide scanning across remote hosts
- QuoxFlow - trigger scans from workflow automations
- QuoxTerminal - context-aware suggestions from the index
- QuoxMEMORY - structural knowledge persists across sessions
- Claude Code - /context-load reads CODEBASE_MAP.md
- CI/CD - scan on push, fail on drift thresholds
# Codebase Mirror - quox-dashboard Generated: 2026-03-04T14:30:00Z Type: React + Express monorepo ## Components (47) - pages/ (12): LandingPage, StorePage, ... - layouts/ (8): DocsLayout, WebsiteLayout, ... - widgets/ (27): ProductRow, SidebarNav, ... ## Routes (34) / → LandingPage /store → StorePage /solutions/* → [9 promo pages] /docs/* → DocsLayout → [25 pages] ## Registries - pluginRegistry: 14 plugins - entityRegistry: 8 entity types - sidebarConfig: 6 sections ## Exports Named: 156 | Default: 42 | Re-exports: 23
Install the scanner
The Codebase Mirror scanner is available as a plugin for indexing external codebases
The cognitive implant is built into QuoxCORE. The scanner plugin extends it to index any codebase - local repos, remote hosts via QuoxBastion, or CI pipelines. Configure scan schedules, drift thresholds, and output formats.
View Plugin DetailsWhat's built, what isn't yet
- Scan, index and query are real: the collector serves /api/v1/codebase/scan, /:id/query, /:id/diff, /:id/summary and /:id/markdown, backed by scanner, diff and query test suites
- Project-type detection covers five languages by file marker (React, Go, Node, Tauri+Rust, Python); deep registry, route and export extraction is built out for React, Go, Node and Tauri projects
- CODEBASE_MAP.md export is a real endpoint (POST /:id/markdown), not a mock file
- There is no CLI: the real interface today is the collector HTTP API (/api/v1/codebase/scan, /query, /diff) and the dashboard’s Codebase Mirror view
- Structured natural-language query is thin: the request shown above is a real call against this scan and it came back with zero results today
- "Scheduled scans" (commit hooks, cron, CI) and "Remote fleet scanning via QuoxBastion" describe no code found in the scanner: a scan runs on demand from the dashboard today, not on a schedule or against a remote host
- Python and Rust projects are detected but fall back to generic-scanner depth — they don’t yet get the registry/route extraction React and Go projects do
Stop reading files. Start querying structure.
One scan. One index. Every AI prompt in your stack gets structural awareness for ~30 tokens. No more guessing which files matter.
How you get it
Free in the store, not shipped yet
Codebase Mirror costs nothing, and the catalogue carries it as coming soon: the row is there, the build is not out. It is the map the rest of the Dev Suite reads.
On its own
Regenerates a live map of a repository from the actual code: registries, routes, imports and configs, written back as CODEBASE_MAP.md so an agent orients from fact rather than from a stale README.
Ask it questions
The map tells you what exists. RepoBrain is what answers "where is X and what imports it" with cited snippets instead of whole files, at a fraction of the context cost.