Docs / The Command Deck

[ PRODUCT TOUR ]

The Command Deck

Six screens, one job each. This page says what every one of them shows, which parts are backed by real data, and what you can actually change from it.

Getting in

The deck lives behind /app and every route under it requires a session. Arrive without one and you are redirected to /login; the deck is also marked noindex, since a signed-out crawler would only ever see the login screen.

The header shows the address your session is scoped to and whether you are in wallet or demo mode. That address is the boundary for every query the deck makes.

Command Deck — the overview

Landing screen, at /app. Four regions:

  • Stat strip — portfolio value, active watches, pending approvals, and fleet state at a glance.
  • Pending approvals — the drafts waiting on you, which is the only thing on this screen that needs action.
  • Live feed — a rolling status stream. Presentation; it is not a log, and the ledger is.
  • Portfolio table — holdings priced at the latest quotes, with a badge naming the price source.

Watches

At /app/watches. The form on this screen is where a voyage starts: symbol, threshold, action, and an optional note passed to the agent when it drafts. The assigned agent is shown before you submit — the same value the server stores, so what you saw is what took the job.

The table below it gives you three controls per watch:

CONTROLEFFECT
Pause / resumeA paused watch is skipped by every evaluation, scheduled or manual.
Evaluate nowRuns the sweep immediately for your watches only — identical code to the cron path, so it is a real test rather than a simulation of one.
Last triggeredWhen this watch last fired. Within 30 minutes of that stamp it is in cooldown and will not fire again.

Proposals

At /app/proposals. Each card carries the agent that wrote it, the kind of move, the rationale, and a current-versus-proposed allocation table with the notional the change would move.

Two buttons: approve and reject. There is no third. Marking something executed is not a decision the deck can make, because Oddysey does not place orders — an execution is reported by whoever filled it, over MCP.

Fleet

At /app/fleet. Agent cards with roles, models, and running tasks.

Ledger

At /app/ledger. The append-only audit trail, newest first, with the event kind, the actor, and a detail line. This is the screen to open when you want to answer “who did this, and when” — every drafted plan, every human decision, and every reported fill lands here, and nothing removes rows from it.

Settings

At /app/settings. Six panels, and it is worth knowing which are wired to something:

PANELWHAT IT DOES
MCP accessLive. Issue and revoke deck tokens. A new token’s plaintext is shown once and never again.
NetworkChain and RPC details for the network the deck reports against.
GuardrailsApproval and limit settings, including the human-approval switch.
Model keysBring-your-own provider keys.
NotificationsWhich events you want to hear about.
BillingPlan and daily agent-hour usage.