Contorium

Persistent Memory Layer for AI Coding

Contorium gives AI a continuous understanding of your workspace, goals, Git changes, and coding sessions.

Instead of losing context every chat, model switch, IDE restart, or session recovery,
Contorium continuously maintains a structured workspace memory for AI.

Built for developers who work with AI every day.

Why Contorium?

Most AI coding tools forget everything.

You switch:

• chats
• models
• sessions
• machines
• IDE windows

…and your AI assistant loses track of:

• what you were building
• which files mattered
• recent Git changes
• project intent
• active debugging context

Contorium fixes that.

It continuously tracks your active workspace and turns it into a persistent AI memory layer.

So your AI can continue where you left off.

What Makes Contorium Different?

Contorium is not another AI chat panel.

It is a:

Workspace-aware AI Memory System

Contorium continuously maintains:

• current focus
• active files
• recent edits
• Git changes
• workspace intent
• event history
• session summaries
• compressed context memory

All stored locally inside your workspace.

Designed for Real AI Coding Workflows

Resume AI coding sessions

Close Cursor today.

Open it tomorrow.

Your AI still understands:

• what you were working on
• which files mattered
• what changed
• your coding goal
• recent project activity

Keep AI aware across large codebases

Contorium continuously tracks:

• working-set files
• recent activity
• Git changes
• active editors
• project focus

Perfect for:

• monorepos
• long refactors
• debugging sessions
• AI agent workflows
• large enterprise projects

Switch models without rebuilding context

Move between:

• GPT
• Claude
• Gemini
• DeepSeek

without rebuilding workspace memory every time.

Reduce AI token usage

Large projects can quickly explode token costs.

Contorium reduces unnecessary AI context by:

• tracking only active workspace changes
• prioritizing important files
• compressing recent activity
• filtering noisy paths
• generating compact structured memory

Instead of sending your entire repository every session, Contorium helps AI focus on what actually matters.

Especially useful for:

• long AI coding sessions
• monorepos
• expensive frontier models
• agent loops
• high-frequency AI workflows

Designed to reduce:

• token usage
• repeated context rebuilding
• unnecessary AI calls
• AI cost overhead

Core Features

AI Workspace Memory

Persistent workspace state:

• current task
• notes
• active files
• recent activity
• Git changes
• workspace intent
• session memory

Stored locally inside:

.contora/

Current Focus Tracking

Set your current focus manually.

Contorium continuously updates surrounding workspace context automatically.

Example:

Current Focus:
Refactor payment retry system

AI inferred goals:
- improve retry stability
- optimize error classification
- reduce duplicate requests

Git-aware Context

Contorium automatically tracks:

• staged files
• modified files
• working-tree changes

and prioritizes them in AI context generation.

Context Compression

Large codebases generate noisy AI context.

Contorium compresses workspace activity into:

• semantic summaries
• ranked priority files
• compact event history
• structured memory blocks

Designed for long-running AI workflows.

Workspace Intent Analysis

Analyze project direction using optional BYOK AI providers.

Generate:

• inferred goals
• workspace intent
• feature direction
• task grouping

Results are stored locally and reused across sessions.

Session Recovery

Save and restore:

• open editors
• workspace state
• active memory
• project context

Your AI coding session becomes persistent.

Local-first Architecture

Contorium is designed to work locally first.

• No cloud sync
• No chat log scraping
• No hidden telemetry
• Workspace-owned memory

Your workspace memory stays under your control.

Optional BYOK AI Features

Bring your own API keys:

• OpenAI
• Claude
• Gemini
• DeepSeek

Used only when running optional AI commands.

API keys are stored securely in:

• VS Code SecretStorage

Never inside:

• settings.json

How It Works

Workspace Activity
        ↓
Workspace Scanner
        ↓
Memory Builder
        ↓
Context Compression
        ↓
Structured Workspace Memory
        ↓
Export / Restore / AI Workflows

Example Workflow

1. Open your project

Contorium begins tracking:

• active files
• recent edits
• Git changes
• workspace activity

2. Set your current focus

Example:

Refactor payment retry system

3. Work normally

Contorium continuously builds:

• workspace memory
• ranked file priority
• semantic summaries
• event history
• compressed context

4. Export AI context

One click.

Contorium generates:

• structured memory
• compressed workspace context
• AI-ready summaries

for your preferred model or agent.

Sidebar Features

• AI current focus
• AI inferred goals
• Workspace summary
• Active files
• Git changes
• Context notes
• Session save / restore
• Semantic summary
• Workspace intent analysis
• Context compression preview

Use Cases

Long Refactors

Keep AI aware across:

• dozens of files
• multiple sessions
• evolving goals

AI Pair Programming

Give your AI assistant:

• project awareness
• Git context
• active workspace intent
• compact memory

Large Monorepos

Reduce noise using:

• ignore rules
• ranking
• token budgets
• compressed workspace memory

AI Agents & Automation

Generate structured workspace memory for:

• agents
• workflows
• automation pipelines
• external AI tools

Expensive Frontier Models

Reduce unnecessary token usage when using:

• GPT-5
• Claude
• Gemini
• long-context workflows

Sidebar Overview

The Contorium sidebar includes:

AI Status

• current focus
• inferred goals
• model/runtime summary

Workspace Context

• active files
• recent activity
• Git changes
• workspace notes

Session Management

• save state
• restore editors
• session persistence

AI Tools

• semantic summaries
• workspace intent analysis
• compressed context previews

On-Disk Layout

<workspace-root>/
├── .contoraignore
└── .contora/
    ├── state.json
    ├── events/
    ├── last-intent.json
    └── memory/

Privacy & Security

• Local-first architecture
• No cloud workspace storage
• No hidden telemetry
• No session scraping
• BYOK optional
• Full workspace ownership

Installation

VSIX

Extensions → Install from VSIX…

From source

git clone https://github.com/frankleeeeeee/contora.git

cd contora

npm install

npm run compile

Press:

F5

to launch Extension Development Host.

Configuration

Key settings include:

Tech Stack

• TypeScript
• VS Code Extension API
• simple-git
• Workspace scanners
• Local memory builders
• Structured context adapters

Architecture

src/
├── core/
├── ai/
├── scanner/
├── state/
├── ui/
├── storage/
└── env/

Vision

AI coding tools should not lose context every session.

Contorium is building:

Persistent Workspace Memory for AI

The next layer of AI-native development environments.

Roadmap

Future directions:

• AI timeline memory
• multi-session memory
• team workspace memory
• agent memory systems
• workspace knowledge graphs
• smarter context scheduling
• adaptive token optimization

License

MIT License

Built for developers who work with AI every day.