#What the Terminal is

The Terminal is a browser-based command line rendered with xterm.js. It pairs a large registry of built-in commands (which run deterministic, tenant-scoped Supabase queries) with an AI fallback for anything you phrase in plain English. Every command runs against your own workspace only — the server derives your tenant_id from your authenticated session, so you can never query another tenant's data.

• Instant analytics — leads, deals, tasks, members, forecast, status render live stat panels.
• Navigation — go pipeline, home, or view lead John Smith jump you straight to a record or view.
• Revenue & ledger — invoices, quotes, revenue, ledger status pull financial intelligence.
• Voice control — say a wake word ("computer", "acc") and speak a command instead of typing.
• Natural language — anything not matched as a built-in is passed to the AI assistant.

#Opening and driving the console

1. 1
   Open the Command Center

   Navigate to /dashboard/terminal, or from anywhere in the app type go terminal. The keyboard shortcut ⌘J is reserved for opening the Terminal.
2. 2
   Type a command

   Start with help to see every command grouped by area (Starship, Analytics, Navigation, Quick Actions, Search, Communications, Revenue, Accounts, Landing Pages, System, AI Mode).
3. 3
   Use autocomplete and history

   Press Tab to complete a partial command; a single match completes inline, multiple matches list as suggestions. Use ↑ / ↓ to scroll through your command history.
4. 4
   Read the result panel

   Successful commands print a short confirmation line and open an animated overlay panel. Press ESC to close the panel, or minimize it to a dock at the bottom of the screen.

Voice commands

The Command Center header includes a voice interface built on the browser's Web Speech API. Say a wake word — the defaults are "computer", "acc", and "hey acc" — then speak a request such as "show me today's leads". The transcript is parsed into a terminal command and executed, and (when enabled) the assistant speaks a short acknowledgement back. Voice requires a secure context (HTTPS or localhost) and a browser with speech recognition support.

#How a command is resolved

When you press Enter, the raw text is sent to POST /api/terminal/execute, which calls executeCommand() in the command registry. Resolution follows a fixed order — the first matching stage wins:

1. Exact match — the first word matches a built-in command name or one of its aliases (e.g. deals, or its alias pipeline).
2. Prefix match — if the first word is at least 3 characters, it matches a command whose name or alias starts with it (so fore resolves to forecast).
3. Revenue commands — invoices, quotes, payments, and revenue reports are checked next.
4. Ledger commands — financial-intelligence commands like ledger status are checked after revenue.
5. Status heuristics — free text containing "ready", "status", or "how are we" runs the status command.
6. AI fallback — anything still unmatched returns passToAI: true, marking it for natural-language handling.

#Command reference

Built-in commands are read-first: they query your workspace and render a panel rather than silently mutating data. Below are the most-used commands grouped by area.

❯ deals
Pipeline: $2.4M total, $1.1M weighted

❯ view deal Acme Renewal
Opening Acme Renewal...

❯ stage Acme Renewal to Proposal
Ready to move "Acme Renewal" to Proposal

❯ forecast
Weighted forecast: $1.1M

#Natural language and the AI assistant

When no built-in matches, executeCommand returns passToAI: true. Conversational AI itself is served by a dedicated streaming endpoint, POST /api/ai/assistant, which the in-app AI Assistant panel uses. It is powered by Anthropic's Claude (model claude-sonnet-4-20250514) via the @anthropic-ai/sdk package and streams tokens back as Server-Sent Events.

The assistant is primed with an Olyron CRM system prompt so it answers in the product's terms — leads, contacts, deals, accounts, pipeline stages, blueprints, cadences, and landing pages. If you pass a context_type and context_id (contact, deal, or lead), the server injects a short snippet about that record into the prompt so replies are grounded in the record you're looking at. Conversations are persisted to the crm_ai_threads table so a thread can be continued via its thread_id.

curl -N -X POST https://app.olyron.com/api/ai/assistant \
  -H "Content-Type: application/json" \
  -d '{
    "message": "Draft a renewal follow-up for the Acme deal",
    "context_type": "deal",
    "context_id": "3f9a...c2e1"
  }'

# Response is text/event-stream:
# data: {"text":"Here's a draft"}
# data: {"text":" follow-up..."}
# data: {"done":true,"thread_id":"9b2e...44af"}

#The execute API

Every terminal command flows through one endpoint. It authenticates the caller, looks up their tenant_id from the profiles table, and runs the command with that tenant scope.

POST /api/terminal/execute

commandstring (required)

The raw command text, exactly as typed (e.g. "leads" or "open deal Acme"). Must be a non-empty string or the endpoint returns 400.

The server ignores any tenant or user identifiers sent in the body and instead derives them from the session, so the endpoint can't be used to reach another workspace. Responses have this shape:

successboolean

Whether the command executed successfully.

outputstring

A short human-readable result line (empty for AI-bound commands).

errorstring | undefined

Present only when success is false.

panelobject | undefined

Structured panel data for the animated overlay.

// Request
{ "command": "leads" }

// Response
{
  "success": true,
  "output": "Found 128 total leads",
  "panel": {
    "id": "a1b2c3d4-...",
    "type": "leads",
    "title": "Leads Analytics",
    "subtitle": "6 new today, 21 this week",
    "icon": "leads",
    "data": { "stats": { "total": 128 }, "byStatus": {}, "bySource": {} },
    "timestamp": "2026-07-10T14:22:00.000Z"
  }
}

#Limits & troubleshooting

There is also a separate, hidden Olyron mesh visualization at /crm/olyron, gated behind the NEXT_PUBLIC_OLYRON_ENABLED feature flag. It is not linked from navigation and is unrelated to the everyday Command Center — if you see "Olyron is not enabled", that flag is simply turned off.