Contox: AI Context Memory
Give your AI tools persistent memory across sessions.
Contox captures your project's architecture, conventions, and decisions so all your AI assistants always have the right context. No more repeating yourself.
One extension, all your tools. Contox auto-configures MCP (Model Context Protocol) for:
| Tool |
Config location |
| Claude Code |
.mcp.json |
| Cursor |
.cursor/mcp.json |
| GitHub Copilot |
.vscode/mcp.json |
| Windsurf |
~/.codeium/windsurf/mcp_config.json |
| Antigravity |
~/.gemini/antigravity/mcp_config.json |
| Cline |
VS Code globalStorage |
| Gemini CLI |
~/.gemini/settings.json |
All configs are written automatically when you connect. No manual editing needed.
Features
Automatic Context Capture
The extension watches your git commits and file saves, building a rich memory of your project over time. Every meaningful commit is enriched with AI-generated summaries.
Smart Connect Flow
Click "Connect IDE" from the dashboard and the extension:
- Checks if the extension is already installed and up-to-date
- Detects if the IDE is already connected
- Auto-creates an API key if needed
- Deploys the MCP server binary
- Configures all 7 AI tools at once
- Starts capturing immediately
If something breaks, use "Reconnect" to re-verify and repair all configs.
MCP Server (Bundled)
The extension bundles and auto-deploys an MCP server that exposes tools to your AI assistant:
| Tool |
What it does |
contox_get_memory |
Load full project context at session start |
contox_save_session |
Save work into categorized sub-contexts |
contox_list_findings |
Browse memory items by category (security, architecture, bugs, etc.) |
contox_search |
Semantic search across project memory |
contox_ask |
Natural language questions about your project |
contox_context_pack |
Get focused context for the current task |
contox_scan |
Auto-extract architecture from codebase |
contox_git_digest |
Read git commits since last save |
contox_hygiene |
Clean up and organize memory |
Genesis Scan
Auto-extract architecture, conventions, security patterns, and data flow from your codebase. Results feed directly into your project memory.
Smart Context Injection
The extension auto-injects Contox instructions into all detected AI agent config files:
CLAUDE.md (Claude Code) - always created.cursorrules (Cursor).windsurfrules (Windsurf) - always created.antigravityrules (Antigravity).clinerules (Cline).github/copilot-instructions.md (GitHub Copilot)
MCP-capable agents get protocol instructions; non-MCP agents get file-based instructions. When available, a project brief from the brain summary is included. Config files are injected on every connect and reconnect.
Active File Context
When you switch files, the extension searches your project memory for relevant items and writes a focused context file at .contox/context.md. Search uses composite scoring with file-overlap boosting for better relevance.
Version Tracking
The dashboard detects your installed extension version and shows an update banner when a new version is available.
Quick Start
- Sign up at contox.dev
- Install this extension in VS Code, Cursor, Windsurf, or Antigravity
- Click "Connect IDE" from the dashboard -- it handles everything
- Done. Your AI assistant now has persistent project memory.
How It Works
You code normally
|
Contox Extension captures git commits + file saves
|
Events are signed (HMAC-SHA256) and sent to Contox API
|
AI enrichment extracts architecture, conventions, decisions
|
MCP server exposes memory to Claude, Cursor, Copilot, etc.
|
Your AI assistant knows your project across sessions
Commands
| Command |
Description |
Contox: Setup Wizard |
Guided setup with API key, team, project, and AI tool selection |
Contox: Load Memory |
Load project memory into the sidebar |
Contox: Sync Contexts |
Manually sync contexts |
Contox: End Session & Start New |
Close current session and start fresh |
Contox: Disconnect |
Pause sync without removing config |
Contox: Reconnect |
Resume sync |
Requirements
- VS Code 1.85+ (or Cursor, Windsurf, Antigravity)
- Node.js 18+ (for MCP server)
- A free account at contox.dev
Privacy & Data Collection
What data is collected
- Git commit metadata: commit SHA, message, author name, timestamp, list of files changed
- Code diffs: truncated to max 3KB per commit (can be disabled in settings)
- File save events: file paths of saved files (no file contents)
- Project metadata: project name, workspace root path
- Session data: session start/end times, event counts, extension version
What is NOT collected
- Full source code or file contents (only truncated diffs)
- Files matching exclude patterns:
.env, .key, .pem, .p12, .pfx, node_modules/, .git/, dist/
- Any data from outside the connected workspace
Purpose
Data is used to build a persistent project memory that AI tools can query via MCP. This includes generating context summaries, semantic search indexes, and session enrichment.
Data sharing
- Data is transmitted to
contox.dev over HTTPS
- Requests are signed with HMAC-SHA256 for tamper protection
- Data is accessible only to your team members (role-based access control)
- Data is not shared with third parties, sold, or used for advertising
Storage & retention
- All data is stored in the EU (Frankfurt region) on Appwrite Cloud infrastructure
- Data is retained for the lifetime of your account
- You can delete sessions, contexts, or your entire project from the dashboard
- Account deletion removes all associated data
User controls
contox.capture.enabled: Disable all event capture (default: enabled)contox.capture.includeDiffs: Disable code diff capture (default: enabled)contox.capture.anonymizeDiffs: Strip code content, keep only file paths and stats (default: disabled)contox.capture.excludePatterns: Customize file exclusion patternscontox.autoEnrich: Disable automatic learning from commits (default: enabled)Contox: Disconnect command: Pause all sync without removing configuration
Links
License
MIT
