The AI coding ecosystem is fragmenting.
Some developers prefer Cursor.
Others use Claude Code.
Many are experimenting with Gemini CLI, Codex, and MCP-powered workflows.
The challenge isn’t choosing the best model.
The challenge is maintaining a consistent understanding of the project while moving between tools.
The Context Problem
Most AI assistants rely heavily on:
- Chat history
- Context windows
- Prompt engineering
These approaches work well inside a single session.
They break down when developers switch tools.
Project understanding becomes fragmented.
A Different Approach
Instead of treating project understanding as conversation memory, we’re experimenting with a different model:
A shared workspace state layer.
The workspace itself becomes the source of truth.
Structured state is generated from workspace activity and stored alongside the project.
Different tools can then consume that state through:
- IDE integrations
- MCP servers
- CLI workflows
Multi-Session Project Understanding
One concept we’re exploring is the idea of session views.
An IDE window, an MCP agent, or a CLI process are not separate memory systems.
They’re simply different views over the same workspace state.
This allows project understanding to remain consistent even as tools change.
Current Direction
The latest Contorium architecture focuses on:
- Shared workspace state
- Cross-tool continuity
- MCP compatibility
- IDE independence
- Multi-session workflows
The long-term goal is straightforward:
Project understanding should be portable.
Not tied to a single editor.
Not tied to a single model.
Not tied to a single conversation.
If AI development becomes increasingly multi-tool, we believe the underlying project state should be shared.
That’s the direction we’re building toward.
More info:https://www.contorium.dev
https://github.com/ContoriumLabs/contorium