Command Agent

Command is a Grok-powered agent with read-only access to your terminal. Ask a question in plain English and it pulls the data, runs the analysis, and answers in context — without making you click through tabs.

Where to find it

Two surfaces, one agent:

  • Command bar — a slim input docked at the bottom of every authenticated page. Press the keyboard shortcut or click in to expand the conversation drawer over whatever tab you're on. Useful for one-off questions while you're working.
  • /command page — a full-screen view with conversation history, prompt suggestions, and side-by-side tool calls. Useful for longer back-and-forth.

Both share the same active conversation, so you can start a question in the command bar on the Risk tab and continue it on the /command page without losing context.

What it can do

The agent has access to roughly thirty internal tools — the same registry exposed to external agents through the MCP server. It can answer questions like:

  • "What's my current net delta in tech?"
  • "Show me my biggest concentration risks above 8%."
  • "Which holdings have earnings in the next two weeks?"
  • "What's AAPL's forward P/E vs. the rest of my portfolio?"
  • "Summarize the latest 8-K from NVDA."
  • "What's the X sentiment on TSLA today?"
  • "Pull the option chain for SPY 60 days out around 480-520 strikes."
  • "Group my positions by basket and show net delta per basket."

It can read positions, balances, margin, Greeks, factor tilts, sector and concentration breakdowns, CME schedules, fundamentals, options chains, financial statements, price history, corporate actions, SEC filings, X news, Grok sentiment, baskets, and your saved Grok and X timeline searches.

Market data — your broker, your entitlements

When you ask the agent for a quote, a chain, fundamentals, or price history, the data comes from your own connected broker under your own market-data entitlement — not from a fan-out third-party feed shared across our user base. If you have Schwab connected, your equity quotes, option chains with native Greeks, and fundamentals come from Schwab's market-data API authenticated with your token. If you don't, the agent falls back to a public delayed source so you still get an answer.

What that means in practice:

  • Real-time when you're entitled. Schwab-connected users typically get real-time quotes for equities, options, futures, and indices — including pre-market and after-hours where the broker provides it. The exact freshness depends on your account's entitlements with Schwab, not ours.
  • Native Greeks on chains. Option chains pulled from Schwab carry broker-supplied delta/gamma/theta/vega/IV per contract. No client-side approximation. The fallback path computes Greeks via Black-Scholes when the broker doesn't provide them.
  • Provenance visible. Every market-data answer the agent returns carries a source and a delayed flag. Ask the agent and it'll tell you where the data came from — schwab, yfinance (delayed), or whichever path served the request. The Risk → Factor Tilts panel surfaces the same field per row in its breakdown table so you can see whether each holding scored from broker-grade fundamentals or public reference data.
  • No cross-user leakage. Cache keys are scoped per user, so your broker's data never serves another account.

You don't need to know any of this to use the agent — ask a normal question and you get an answer. But if you ever wonder why a quote feels fresher than a public site, or why AAPL's fundamentals look more current than what you'd get from a free aggregator, that's the broker pipe under the hood.

Examples

  • "What's AAPL's last? Tell me whether it's real-time or delayed."
  • "Pull the SPY chain — strikes within 5% of spot, next two expirations."
  • "What was Apple's price action this past week?"
  • "Show me MSFT's PE, ROE, and operating margin alongside the market-cap-weighted averages for my portfolio."

Symbol smarts

You can paste a symbol in any common notation and the agent will figure out what you mean. The Command Agent sits on a cross-platform symbol resolver that knows the same instrument has different names depending on the venue:

  • MGCM26, /MGCM26, GC=F → June 2026 micro gold futures
  • ESM26, /ESM26, ES=F → June 2026 E-mini S&P
  • SPX, $SPX, $SPX.X, ^SPX → S&P 500 index
  • VIX, $VIX, ^VIX → CBOE Volatility Index
  • AAPL 03/21/26 $200C → Apple March 2026 $200 call

What the resolver handles for you:

  • Futures contracts in either broker notation (with or without the leading slash) or Yahoo-style continuous symbols.
  • Indices across the $SYM / $SYM.X / ^SYM conventions used by different brokers and data sources.
  • Micro vs. full contracts — when you ask about "gold," the agent picks the right root for your context (MGC for retail-sized exposure, GC for the full-size contract).
  • Cashtag normalization for X/Twitter searches — the agent automatically uses $GC (the social-media cashtag) when searching X, even if you typed MGCM26.

If you give it something genuinely ambiguous ("oil"), it'll ask which contract you mean. If you give it something nonsensical (XYZQ99), it'll say so rather than silently returning empty data.

Examples

  • "What's gold doing? Pull MGCM26." — agent fetches the broker quote and history for /MGCM26.
  • "What's the chatter on $ES today?" — agent translates the cashtag to the right futures root for X search.
  • "Compare /NQU26 to /ESU26 over the last month." — both translate cleanly to the broker's price-history endpoint.

What it can't do

The agent is read-only by design. It will never place a trade, modify a position, change a setting, or move money. The only "write" actions it can perform are three explicit triggers, each of which costs from your account's credits or quotas:

  • Trigger filing analysis — queue a fresh Grok analysis on a SEC filing.
  • Trigger Grok Search — run a new persistent Grok search (counts against your daily Grok Search credit).
  • Trigger X Timeline search — run an X timeline search on your account lists. Requires explicit confirmation in the chat ("yes" / "go ahead") before it runs, so you don't burn allocation by accident.

Anything else stays in the read-only band.

Tiers and limits

Each conversation message you send is a "turn." Turns are metered per day, resetting at UTC midnight:

  • Free — Command is locked. Upgrade to use it.
  • Standard — 10 turns per day.
  • Premium — 30 turns per day.

A turn covers the entire round-trip including any tool calls the agent makes — so one question that pulls Greeks, then options chains, then sentiment is still one turn. Your remaining turns and reset time are visible in the command drawer header.

Built-in Grok features web_search and x_search run server-side at xAI as part of the agent's reasoning and don't separately count against your Grok Search credits.

Tips

  • Be specific about scope. "What's my exposure?" is ambiguous. "What's my delta-weighted exposure to semis?" gets a precise answer.
  • Combine in one turn. Ask "what are my top 5 concentration risks and their next earnings dates" rather than splitting into two — same turn cost.
  • Use baskets. Once you've grouped positions into thematic baskets, you can ask the agent to slice analysis by basket: "show me the Greeks breakdown for my AI basket."
  • Trust the citations. When the agent runs a Grok Search or web search, it returns the sources used. Click through to verify anything before acting on it.

Command vs. MCP

Both surfaces use the same underlying tool registry but serve different needs:

  • Command Agent — conversational, in-app, optimized for human use. Daily turn quota.
  • MCP Server — programmatic, for connecting external AI agents (Claude Desktop, custom agents) to the same data. Per-tool calls, no turn quota — only the metered tools cost.

If you're querying interactively while you trade, use Command. If you're building automation around your portfolio, use MCP.