The cockpit for your local-LLM team. Live.

Without this panel, your orchestrator knows what's happening — and you don't.

The main opencode-agent-intercom plugin injects a live subagent snapshot into the orchestrator's hidden system prompt. You only see it if you stop and ask. After a while, you stop asking. You fly blind.

This plugin puts that snapshot — and every runtime knob worth tuning — right in your opencode sidebar.

┌─ ▼ Subagents (2) ──────────────────────────┐
│   coder#1      busy   6.2 K   0:42   [✕]   │
│   designer#1   idle   2.1 K   1:08   [✕]   │
└────────────────────────────────────────────┘
┌─ ▼ Limits ─────────────────────────────────┐
│   max subagents   [-]   3   [+]            │
│   max Token(k)    [-]  40   [+]            │
│   thinking        [on]                     │
│   tool details    [on]                     │
└────────────────────────────────────────────┘
┌─ ▼ LLM params    [<] orchestrator  [>] ────┐
│   temperature    [-]  0.70 [+]   ★         │
│   top_p          [-]  0.90 [+]             │
│   top_k          [-]    40 [+]   ★         │
│   min_p          [-]  0.05 [+]             │
│   repeat_penalty [-]  1.05 [+]             │
│                                            │
│   [reset current agent]                    │
└────────────────────────────────────────────┘

• See every subagent, always. Name, status, context size, age — live. No more "what's the coder doing right now?" detours through the orchestrator.

• Open a subagent's full transcript in one click. Watch its work. Stuck? Hit ✕ to abort. The orchestrator is told why and re-plans.

• Tune the framework live. max subagents, max Token(k), thinking and tool-details visibility — change them, the running plugin picks it up in ~2 seconds. No opencode restart.

• Sample your model per role. Different temperature for the orchestrator, your coder, your designer. Plus llama.cpp specifics (min_p, repeat_penalty, chat_template_kwargs) routed through output.options. A ★ marks an override; [reset current agent] drops back to the role's default. Applies on the next LLM call. Find the sweet spot for your model without leaving opencode.

• Stay oriented. When a subagent finishes and its session vanishes, the panel drops you back into the orchestrator chat — not the home page.

Every change is written to a file under ~/.config/opencode/. The main plugin watches. There is nothing else to wire.

Mouse

• Click a subagent's name → opens its session.
• Click the trailing ✕ → aborts that subagent.
• Click any ▼ section header → collapse / expand.
• Click any [-] / [+] → adjust value (hold for hot-repeat).

Keyboard (focus the panel first — Alt+A or click the header)

The installer for the main plugin (npx opencode-agent-intercom-install, run from your project) adds this TUI plugin automatically. Restart the opencode TUI to see the panel.

Manual fallback — TUI plugins are registered in ~/.config/opencode/tui.json (user-global, not opencode.json):

{
  "plugin": ["opencode-agent-intercom-tui"]
}

Local checkout? Point at the built file directly — opencode does not resolve a TUI plugin from a directory path:

{
  "plugin": ["/path/to/opencode-agent-intercom/tui/dist/tui.js"]
}

Run npm run build here first.

npm install
npm run build      # tsup → dist/tui.js
npm run typecheck

Runtime deps (@opentui/*, solid-js, @opencode-ai/plugin) are provided by opencode and marked external in the build. They live as devDependencies here only — adding them as peerDependencies would shadow opencode's own copies at runtime and break loading.

MIT — see LICENSE.