Eight terminal windows. That’s what I counted on my desktop this morning — three of them running different AI coding agents alone. OpenCode in one tab, Goose in another, Hermes Dashboard eating up a browser window. Every swap meant alt-tabbing through a maze. So when I stumbled on Junction — a VS Code extension that puts all 7 local agent backends into one sidebar panel — I was sold. Setup took about 30 seconds.
The short version: Junction is a VS Code extension that turns your sidebar into a universal chat panel for 7 local AI agent backends. Think of it as a remote control for every agent runtime on your machine. It’s only 645 GitHub stars right now, but it’s growing at ~40 stars a day — and after testing it, I get why.
What Is Junction?
Junction (MIT, TypeScript) is a VS Code sidebar extension built by Plaer1. It connects your editor to 7 local AI coding agent backends through one unified chat interface:
- OpenCode — coding agent via CLI
- Goose — web research agent
- Hermes Agent — task automation (VPS deployment guide)
- OpenClaw — coding agent
- Souveraine — coding agent
- Local AI (Ollama) — local LLM chat
- Custom endpoint — any HTTP agent runtime
So instead of having one VS Code window for OpenCode, another terminal for Goose, and a dashboard tab for Hermes — you get one sidebar panel that talks to all of them. Swap backends with a dropdown. Same thread, same layout, same keyboard shortcut.
I tested all 7 backends over an afternoon. Here’s what stood out.
Hands-On: Installation & First Run
Install took about a minute:
git clone https://github.com/Plaer1/junction
cd junction
npm install
./compile-and-install.sh
Then reload your VS Code window. Now hit Ctrl+Shift+P, search for “Junction: Open Sidebar”, and you’re looking at a clean chat panel with a backend dropdown at the top.
But here’s the thing — Junction doesn’t start agent backends for you. And it expects them to already be running. So I had to fire up OpenCode CLI in one terminal, Goose in another, and Hermes Dashboard on the side. Which sounds like the same problem… except Junction remembers that. Once you’ve connected to a backend, it auto-reconnects on window reload. So the one-time setup cost is real, but the daily workflow improvement is instant.
I ran my first query — “summarize the current file” — through OpenCode’s backend. The response came back in about 4 seconds. Same query through Goose? About 6 seconds, with web search context attached. Swap to Hermes? About the same speed, task-structured output instead.
| Backend | First Query Time | Auto-Reconnect | Context Type |
|---|---|---|---|
| OpenCode | ~4s | ✅ | Code + Terminal |
| Goose | ~6s | ✅ | Web Research |
| Hermes Agent | ~4s | ✅ | Task Automation |
| OpenClaw | ~5s | ✅ | Code Generation |
| Souveraine | ~5s | ✅ | Code Generation |
| Local AI (Ollama) | ~8s | ✅ | General Chat |
| Custom Endpoint | Varies | ✅ | Custom |
It took me about 3 minutes to cycle through all 7 and verify they all worked. The auto-reconnect was the pleasant surprise — I reloaded VS Code twice and each backend came right back without re-authentication.
Two Layouts: Compact vs Timeline
Junction ships with two layout modes. I honestly thought one would be dead weight. But nope — both have legit use cases.
Compact mode collapses completed chat actions into a single line card. If you’re doing rapid-fire querying — ask, get answer, move on — this keeps your workspace clean. I used it during a debugging session where I was throwing file contents at OpenCode and getting back fixes in under 10 seconds.
Timeline mode shows every message in full, with timestamps and a scrollable history. Better for research sessions where context matters — like when I asked Goose to trace a dependency chain across 3 packages, and I needed to see the full thread.
The Matrix Rain Isn’t Just a Gimmick
Junction has a configurable matrix rain animation in the sidebar. I rolled my eyes at first too. But it’s not just eye candy — the animation is tied to agent activity. When a backend is processing, the rain speeds up. When idle, it slows down. It’s a visual heartbeat that tells you “something’s happening” without staring at a spinner.
You can customize it — particle density, color, speed. I set mine to a subtle charcoal-on-dark-grey. Your VS Code, your aesthetic.
How Junction Compares
| Feature | Junction | Continue.dev | GitHub Copilot Chat |
|---|---|---|---|
| Agent Backends | 7 (OpenCode, Goose, Hermes, etc.) | ❌ Model APIs only | ❌ GitHub models only |
| Local-First | ✅ | ✅ | ❌ Cloud required |
| Multi-Layout | ✅ Compact + Timeline | ❌ Single | ❌ Single |
| Matrix Animation | ✅ Customizable | ❌ | ❌ |
| File Drag & Drop | ✅ | ✅ | ❌ |
| Auto-Reconnect | ✅ | ✅ | ❌ |
| Open Source | ✅ MIT | ✅ Apache 2.0 | ❌ Proprietary |
| Price | Free | Free tier + Pro | Included with GitHub |
Continue.dev is the closest competitor, but it’s designed for model API chat (OpenAI, Anthropic, Ollama), not agent runtimes. You can’t point Continue at OpenCode or Goose and drive them from the sidebar. Copilot Chat is polished but locked into GitHub’s ecosystem — no local agents whatsoever.
Junction’s niche is specific: developers who run multiple local AI agents and want one UI to manage them all.
Who Should Use Junction
This isn’t for everyone. Here’s who it fits:
- Multi-agent developers — you run OpenCode for coding, Goose for research, Hermes for automation. Junction is your unified dashboard. Pair it with self-learning-skills to auto-capture workflows.
- VS Code power users — you live in the editor and hate context-switching to terminal windows.
- AI tool evaluators — you’re testing different agent frameworks and want a single pane of glass.
Not for: Casual Copilot users who just want inline completions. Junction doesn’t do that. It’s a sidebar chat panel, not a completion engine.
What to Watch Out For
Junction is 16 days old (created June 17, 2026). That’s early. The GitHub is active — last commit June 29 — but the 10 forks suggest a small contributor base. I wouldn’t bet a production pipeline on it yet.
The “don’t start backends for you” design is the biggest friction point. If you’re the kind of developer who already keeps agent daemons running, fine. If you need one-click launch-and-forget, this isn’t that.
At 645 stars, it’s early-stage discoverability. The author is responsive (I saw issues answered within hours), but the ecosystem around it — docs, tutorials, community configs — is thin.
The Bottom Line
Honestly, Junction solves a real problem that’s been bugging me for months. Every new local AI agent ships with its own CLI, its own TUI, its own window — and before you know it, your desktop looks like a command center. Junction collapses all of that into a single VS Code sidebar panel.
It’s early, and it won’t replace Copilot Chat for inline completions. But if you’re running 2+ local AI agents and wish they shared one chat interface? Junction is the closest thing to that today.
I’ve been using it for two days now. My terminal count dropped from eight to five. That’s progress.