Contorium v2 — A Runtime Continuity Layer for AI Coding Agents

Problem: AI tools are stateless by design

Modern AI coding tools (Cursor, Claude Code, Codex, VS Code agents) share one limitation:

They reset context every session.

This leads to:

• repeated explanations
• lost debugging state
• broken multi-tool workflows
• inconsistent architectural memory

Even with large context windows, the system still does not persist state.

⸻

Solution: Contorium

Contorium introduces a runtime continuity layer between AI agents and the developer workspace.

Instead of relying on prompt history, it persists structured workspace state.

⸻

Core Design

Contorium maintains three layers of continuity:

1. Current Focus Layer

Tracks what the developer is actively building.

1. Workspace State Layer

Monitors:

• active files
• git changes
• recent edits
• working set evolution

1. Session Continuity Layer

Restores context across:

• IDE restarts
• model switching
• multi-agent workflows

⸻

Architecture Overview

Contorium acts as a bridge between:

• IDE extensions (VS Code / Cursor)
• MCP-compatible agents
• CLI-based AI tools

It exposes runtime state so any agent can resume where another left off.

⸻

Key Insight

AI coding evolution is not limited by intelligence.

It is limited by state continuity across tools.

Once state becomes portable, AI agents stop being isolated tools and become part of a shared runtime system.

⸻

V2 Direction

Current development focus:

• improving MCP-based state exchange
• refining workspace graph model
• cross-agent synchronization layer
• reducing context reconstruction overhead

⸻

Vision

We are moving from:

stateless AI conversations

to:

persistent AI development runtimes

⸻

Project: https://www.contorium.dev
https://github.com/ContoriumLabs/contorium