im4codes/imcodes

What it solves

IM.codes provides a messaging, memory, and review layer for terminal-based coding agents. It solves the "reach problem" where agent workflows are interrupted when a user leaves their desk, and the "judgment problem" where single-model blind spots lead to unstable answers on complex tasks.

How it works

The system connects AI agents to the computers where work happens via "Controlled Nodes," allowing agents to operate machines with scoped commands and file transfers. It implements a shared memory layer and a managed Model Context Protocol (MCP) tool surface that works across different AI providers (such as Claude Code, Gemini CLI, and GitHub Copilot). It also includes a supervisor that can evaluate agent turns and automatically continue or trigger audit loops.

Who it’s for

Developers who use terminal-based AI coding agents and need to manage them from mobile or web devices, as well as teams requiring cross-provider agent coordination and structured audit pipelines for code delivery.

Highlights

• AI Remote Desktop: Enroll computers as Controlled Nodes for agents to operate directly, with browser-based remote desktop access for humans on Windows nodes.
• Shared Agent Memory: Converts completed agent work into reusable, multilingual semantic memory to avoid log spam and provide durable recall across sessions.
• OpenSpec Auto Deliver: A supervised pipeline that turns spec-driven changes into a delivery run with proposal audits, implementation loops, and automatic module scoring.
• Managed MCP Tools: Provides a standardized tool surface for memory recall, agent-to-agent messaging, and cron scheduling without exposing raw auth tokens.
• Supervised Execution: A supervisor evaluates agent turns at the idle boundary to decide whether to auto-continue, hand back control, or trigger a rework loop.
• Remote Access Tools: Includes a remote terminal, file browser with Git diffs, and a secure localhost preview tunnel for mobile/web access.