What is Contox?
Contox is a persistent AI project memory system. It captures, enriches, and structures development context so that AI coding assistants can understand your project deeply -- across sessions, tools, and team members.
Every time you start a new session with an AI coding tool, it starts from scratch. It does not remember what you built yesterday, which architectural decisions you made last week, or the bugs you fixed this morning. Contox solves this by giving your AI a structured, always-up-to-date memory of your project.
Why persistent memory matters
AI coding assistants are powerful, but they are stateless. Without memory, they:
- Repeat mistakes you have already corrected
- Ignore conventions your team has established
- Miss context about why code was written a certain way
- Waste time re-discovering your project structure every session
Contox eliminates these problems by maintaining a living knowledge base that your AI loads at the start of every session.
Key features
Persistent memory across sessions
Every session's work -- features built, bugs fixed, decisions made -- is saved into structured memory categories. The next time you (or a teammate) start a session, the AI picks up right where things left off.
Works everywhere
Contox is tool-agnostic. It works with:
- Claude Code (via MCP server)
- Cursor (via MCP or CLI)
- Windsurf (via MCP or CLI)
- GitHub Copilot (via CLI)
- Cline (via MCP)
- Aider (via CLI)
- Any AI tool that can read context files or use MCP
AI-powered enrichment
Raw session data goes through an enrichment pipeline that extracts structured knowledge: architecture patterns, coding conventions, implementation details, key decisions, bugs and fixes, and pending tasks. This is not just a log -- it is curated, categorized project knowledge.
Codebase scanning
The contox scan command analyzes your entire codebase and generates approximately 15-20 structured sub-contexts covering your routes, components, libraries, hooks, dependencies, configuration, and documentation. This gives the AI an instant understanding of your project without reading every file.
Hierarchical brain
Memory is organized into a tiered hierarchy. Tier 1 items (architecture, conventions) are always loaded. Tier 2 items (implementation details, decisions) are loaded on demand. This keeps token budgets manageable even for large projects.
How it works
Contox follows a three-stage cycle:
1. Capture
Context is captured from multiple sources:
- Session saves -- When you or your AI finishes a coding session, a summary and structured changes are sent to Contox
- Codebase scans -- The CLI scans your project structure, routes, components, and dependencies
- Git history -- Commit diffs and messages are digested for enrichment evidence
2. Enrich
The V2 ingest pipeline processes raw session data through AI-powered enrichment:
- Extracts factual claims from session evidence
- Verifies claims against actual code changes
- Categorizes knowledge into architecture, conventions, implementation, decisions, bugs, and todos
- Assigns confidence scores and tracks source references
3. Retrieve
When a new session starts, the AI loads the project brain -- a structured markdown document assembled from all approved memory items. This document includes:
- Project architecture and tech stack
- Coding conventions and style rules
- Implementation journal (what was built and when)
- Key decisions with rationale
- Known bugs and their fixes
- Pending tasks and technical debt
Who is Contox for?
Contox is for any developer who uses AI coding assistants and wants better results over time. It is especially valuable if you:
- Work on projects across multiple sessions
- Use more than one AI tool
- Work on a team where multiple developers use AI assistants
- Want your AI to follow project-specific conventions
- Are tired of re-explaining your project structure every session
Architecture overview
Contox consists of three client-side components that connect to the Contox cloud platform:
| Component | Package | Purpose |
|---|---|---|
| CLI | contox-cli | Universal command-line interface for all AI tools |
| MCP Server | @contox/mcp-server | Model Context Protocol server for Claude Code and compatible tools |
| VS Code Extension | contox.contox-vscode | Integrated experience within VS Code |
All three connect to the Contox API at https://contox.dev/api, where your project memory is stored, enriched, and served.
Next steps
Ready to get started? Head to the Quick Start guide to set up Contox in under 5 minutes.