TangleClaw

TangleClaw v3

<p align="center"> <img src="https://github.com/Jason-Vaughan/project-assets/blob/main/tangleclaw-logo.png?raw=true" alt="TangleClaw logo" width="200"> </p>

<p align="center"> <strong>AI coding session orchestrator</strong> — persistent sessions, multi-engine management, methodology enforcement, mobile access </p>

<p align="center"> <code>claude code</code> &middot; <code>codex</code> &middot; <code>gemini cli</code> &middot; <code>aider</code> &middot; <code>openclaw</code> &middot; <code>tmux</code> &middot; <code>pwa</code> &middot; <code>zero dependencies</code> </p>

<p align="center"> <strong>macOS only</strong> (launchd required for service management) </p>

---

You VPN into your dev machine. You SSH in. You navigate to your project directory, fire up an AI coding agent, and start building. Thirty minutes later your VPN hiccups, or your SSH tunnel drops, or your laptop goes to sleep — and the session is gone. The agent's context, your conversation history, everything. There's no way to reconnect. You SSH back in, start over, and re-explain what you were doing.

TangleClaw was built to fix that. It wraps AI coding sessions in persistent tmux processes so they survive network drops, device switches, and reconnects. Close your laptop at your desk, open your phone on the couch, and pick up the exact same session. The agent never knows you left.

What started as session persistence grew into a full orchestration platform. Once you have persistent sessions, you start wanting a dashboard to manage them. Then you want your development methodology enforced as structural rules, not suggestions the agent can ignore. Then you want engine-native config generated automatically so Claude Code, Codex, Gemini CLI, and Aider all get the same instructions without you maintaining four different config files. Then you want port conflict management across projects, mobile access, idle detection, session wrap protocols.

TangleClaw is all of that — a local platform that sits between you and your AI coding agents, accessible from any browser or phone on your network. Project groups with shared markdown docs, per-session launch-mode selection (from fully interactive to full-autonomy sandboxed), file-based session memory that persists across restarts, a one-click mkcert HTTPS wizard, and a universal project-version detection chain are all wired in.

Features

• Persistent sessions — AI engine sessions run in tmux, surviving network drops, device switches, and reconnects. Close your laptop, switch devices, pick up where you left off
• Five built-in engines — Claude Code, Codex, Gemini CLI, Aider, and OpenClaw. Write rules once — TangleClaw generates engine-native config so every agent gets the same instructions
• Launch mode selector — pick a permission mode when you start a session: Interactive (prompt on every action), Accept Edits (auto-accept file reads and edits), Plan Only (read-only, no code changes), Auto (full autonomy with safety classifier), or Bypass (accepts everything — containers/VMs only). Engines without launchModes launch as before; Claude Code ships with all five wired in
• Project groups & shared docs — link related projects into a group, then share markdown documents across them (architecture notes, network diagrams, runbooks) with per-doc locking so two sessions can't step on each other. Shared directories auto-sync .md files on session launch
• Session memory — file-based, per-project memory system at .tangleclaw/memories/ that persists context across AI sessions. The guide is injected into every engine config so Claude, Codex, Gemini, and Aider all read and update it the same way
• Methodology enforcement — pluggable JSON templates define phases, rules, and session behavior. Rules are structural gates, not suggestions. First-class Prawduct integration for governed workflows with independent Critic review
• PortHub built in — central port registry preventing conflicts across all projects. Originally a standalone CLI, now fully integrated into TangleClaw with permanent and TTL leases, heartbeat support, and system-wide conflict detection via lsof
• HTTPS via mkcert, one click — first-run wizard detects mkcert, generates localhost certs into ~/.tangleclaw/certs/, and hot-swaps the server to HTTPS with a restart overlay. Required for OpenClaw Web UI device pairing; optional manual-cert and skip paths also supported
• Dashboard & mobile PWA — manage projects, launch sessions, and interact with AI agents from any browser or phone on your network. Installable on iOS and Android. Archive projects you're not touching, get an update-available pill when a new TangleClaw ships, and see model status for Claude/OpenAI/Google in the session banner
• OpenClaw integration — connect to remote OpenClaw instances via SSH or Web UI mode with automatic SSH tunnel management, and live background process visibility via ClawBridge
• Universal project version detection — every project's current version is resolved through a layered chain: .tangleclaw/project-version.txt (AI-recorded) → CHANGELOG.md → version.json → package.json. The session wrap re-records it, so the dashboard badge stays honest across any project convention
• Zero dependencies — Node.js 22+ stdlib only. No npm install, no build step, no bundler

<details> <summary>All features</summary>