# Entangle: Share Live AI Coding Sessions with a 3-Word Command

> Source: <https://promptcube3.com/en/threads/4951/>
> Published: 2026-08-04 16:00:51+00:00

# Entangle: Share Live AI Coding Sessions with a 3-Word Command

```
entangle link @user
```

That's it. The person you tag gets a real-time view of your terminal, your [AI agent](/en/tags/ai%20agent/)'s context, and can start contributing immediately. It's like VS Code Live Share but built for AI-assisted workflows.

Here's how I've been using it in practice:

## The workflow I actually use

When I'm pair-programming with a colleague on a tricky [Claude Code](/en/tags/claude%20code/) task, instead of screen-sharing over Zoom and explaining what I'm doing, I just run:

```
entangle link @colleague
```

They get a link in their terminal chat or Discord DM. Clicking it drops them into the same session — same files open, same AI context, same conversation history with the LLM agent. They can type suggestions, run commands, or just watch and learn.

## Configuration that matters

The default settings worked fine, but here's what I changed in `~/.entangle/config.yaml`

:

```
session:
  max_participants: 5
  auto_save: true
  transcript_retention: 7d

ai_context:
  include_prompt_history: true
  include_tool_calls: true
  redact_secrets: true

network:
  relay_region: auto  # picks closest server
```

The `redact_secrets`

flag is the one that actually matters — it scrubs API keys and passwords from the shared context before anyone else sees it.

## Real-world gotchas

**Latency**: If the relay server picks a region far from both of us, there's noticeable input lag. The`auto`

region setting usually works, but I've manually set it to`us-east`

when working with teammates on the East Coast.**File sync**: It shares the current working directory, not a git repo. So make sure you're both on the same branch, or at least have committed your changes.**AI context size**: With multiple people contributing, the LLM agent's context window fills up faster. I've had sessions where we hit[Claude](/en/tags/claude/)'s limits after about 45 minutes of back-and-forth.

For anyone doing remote pair programming with AI coding tools, this solves a real friction point. No more "here's my screen, now watch me type" sessions. You hand off control and the AI context just flows naturally to whoever you invited.

[Next How I Track AI Token Spend Across Claude Code, Cursor, and Codex →](/en/threads/4929/)