CortexPath
AI codebase-intelligence platform: point it at a local project and it ingests every file, writes plain-English logic summaries, maps the dependency graph with a "blast radius" view, and lets you chat with the whole codebase through retrieval-augmented generation.
The problem
Joining an unfamiliar codebase means hours of reading files just to learn what talks to what. Generic AI chat tools have no memory of the project, hallucinate file names, and give no picture of how a change ripples through the system.
The solution
CortexPath turns a local folder into a searchable knowledge base. Each file is parsed for its imports and exports, summarized by an LLM, and embedded locally into pgvector. A dependency graph shows the blast radius of any file, and a chat assistant answers questions with retrieval over the ingested code, so answers cite the real project instead of guessing. Per-user quotas and a model fallback chain keep it usable on a free AI tier.
Screenshots
CortexPath landing page with the live architecture-map preview 
Unified sign-in and sign-up page (Better Auth)
Architecture
- WebNext.js 16React 19React FlowTanStack QueryTailwind CSS 4
- MobileExpoReact NativeNativeWind
- APINext.js route handlersBetter Authts-morph
- AIVercel AI SDKGroqTransformers.js
- DataPostgreSQLpgvectorPrisma
- HostingRailway
Highlights
- Folder ingestion in the browser, processed server-side in batches with live progress; ts-morph extracts imports and exports for every file.
- Logic Summaries: per-file AI breakdown of purpose, exports, security notes and architectural role, cached after the first run.
- Enhancement Studio generates ready-to-paste prompts for five goals: security fix, optimization, system design, blast radius and mentor challenge.
- Interactive architecture map (React Flow + dagre layout): click a file and every dependent animates red to show the impact of a change.
- RAG chat with multiple sessions: embeddings are computed locally with Transformers.js (all-MiniLM-L6-v2) and searched with pgvector, no paid embedding API.
- Model fallback chain on Groq (Qwen 3.8 27B, then GPT-OSS 120B and 20B) with per-user daily quotas tracked in Postgres.
- Companion mobile app built with Expo Router, React Native and NativeWind.