IM.codes is a messaging, memory, and review layer around terminal-based coding agents such as Claude Code, Codex, Gemini CLI, GitHub Copilot, Cursor, OpenCode, OpenClaw, and Qwen. It keeps agent sessions reachable from a browser, iPhone, iPad, or Apple Watch — terminal access, file and git views, localhost preview, notifications — and adds a shared memory layer, a managed Model Context Protocol (MCP) tool surface, supervised execution, and cross-provider audit on top of your own infrastructure.
Controlled Nodes turn enrolled computers into machines an authorized agent (or a team of agents) can operate directly through scoped commands, individual file transfer, and typed Computer Use tools; supported Windows nodes also provide browser-based remote desktop with view and control modes. Each node has independent credentials, stays separate from normal server and session lists, and can have execution access revoked by its owner.
Shared Agent Context & Memory materializes only final assistant outputs from completed work into summaries, keeps raw and processed memory local with optional user-scoped cloud sync, and injects relevant past work per message and at session start with cards showing what was recalled and why. Recall uses multilingual embeddings locally and pgvector server-side. A daemon-managed stdio MCP server exposes memory search and write, agent-to-agent messaging, cron scheduling, and alias tools, with every call bound to the current session, project, user, and server.
Supervised execution evaluates each completed turn at the idle boundary against your own supervision instructions and decides to auto-continue, hand back, or enter an audit-and-rework loop. OpenSpec Auto Deliver extends this into an end-to-end pipeline for a change folder: optional spec audit, a task-driven implementation loop, structured module scores for spec, tasks, implementation, tests, and risk, and a PASS, REWORK, or BLOCKED verdict. It never stages, commits, or pushes; the user remains the release owner. A shared test instance runs at app.im.codes, but self-hosting is the recommended path.
Features
- Remote terminal: raw PTY streaming or a structured chat view with parsed tool calls and thinking blocks, from any browser without SSH or port forwarding
- Controlled Nodes: enroll multiple computers for scoped commands, file transfer, and Computer Use, with per-node revocable credentials
- Windows remote desktop: browser and mobile screen view or control on supported nodes, with display and resolution switching and multiple viewers
- Browser automation: Chrome DevTools Protocol-based open, navigate, click, fill, and evaluate, with a loopback CDP endpoint for scripts
- Shared memory: problem-to-solution summaries from completed work, multilingual recall, automatic injection, and a UI to inspect, query, and archive entries
- Managed MCP tools: searchmemory, saveobservation, savepreference, sendmessage, and cron_* tools bound to the runtime identity
- Supervised execution: per-session off, supervised, or supervised_audit modes with fail-closed handback on timeout or invalid output
- OpenSpec Auto Deliver: spec audit, implementation loop, module scoring, rework gates, and fast, standard, strict, and deep presets
- Team audit: structured cross-model review of work before code lands
- Collaborative sessions: share a tab, sub-session, or server as viewer or participant with actor-labelled messages and revocable access
- Private aliases: owner-scoped snippets inserted with ; markers and resolved out of band
- Mobile and watch clients: iPhone, iPad, and Apple Watch apps plus a web app, with push notifications and quick replies
Integrates with
Claude Code
Terminal-based agentic coding tool that reads your codebase and runs tasks through natural-language commands
Codex
Lightweight coding agent that runs in your terminal
OpenClaw
Self-hosted single-operator assistant runtime that links models, tools, and messaging channels through one local Gateway
OpenCode
Open-source AI coding agent that runs in the terminal, an IDE, or a desktop app
