https://overdeck.localhost (or http://localhost:3011 on a plain
install). This page lists every view, where it lives in the navigation, and what it shows.
Find a view
The left column holds four always-visible entries: Project home (shown only when a project is selected), Flywheel (shows “running” while a run is active), Order books, and More, a dropdown labeled “More views” with every other view in four groups:- Operations: Backlog, Board, Pipeline, Awaiting Merge, Agents, AutoPreso
- Infrastructure: Resources
- Observability: Activity, Sessions, Metrics, Costs, Health, Deacon
- System: Knowledge, Skills, Context, Settings, God View
G keyboard shortcut
(press G, let go, then the second key); see Keyboard shortcuts for
the full list.
Views marked Experimental below only appear in the More menu, and only resolve by direct URL,
once Experimental features is turned on.
Home
Home has two modes, Simple and Advanced, picked with a two-button switch in the Home header in either mode. A fresh browser profile opens in Simple; a browser that has used the dashboard before (it already has a last-viewed tab) opens in Advanced. The choice is stored per browser and is not a server setting. Simple Home leads with a headline that counts what needs you (or reads “Hello.”), three stats (being worked on, ready to merge, finished this week), the composer, and the Get set up card.
Below that, sections appear in order: Needs you, Ready to start (only when non-empty), Having trouble
(only when non-empty), Just filed, Working now, Ready to merge, and Finished (up to 8 rows); only Needs you and Working now show when empty. Simple
Home never merges on its own; you click Merge to main yourself.
Advanced Home leads with the title “System briefing”, the composer and the Get set up card, then
summary tiles for Agents working, Paused, Ready to merge, Needs you and Cost today, followed by the
Activity Feed, Workspaces, Knowledge Registry and Projects sections.
Home is the global landing page at /. It is not the same thing as Project home, which is one
project’s pane inside the Command Deck.
The Home composer starts work directly from typed text. Typing anywhere on the page focuses it.
Once you type, a dropdown offers Ask <agent>: (Enter, starts a conversation), Run in terminal:
(Ctrl+Enter, opens a terminal in the project’s Command Deck), a Codex row when Codex is installed
and not already selected, and, in Simple mode only, Talk it through first: (discusses the task
before filing anything). A project chip picks which project the work lands in; its menu ends with
Add a project…. Model and effort pickers sit in the same row.
The Get set up card is titled “Get set up” with the subtitle “A few things make Overdeck work
best. Nothing here stops you from starting.” Its rows cover signing in to Claude, connecting GitHub,
getting agents ready (shown only when the terminal backend is not tmux), installing a terminal
helper, and Docker (optional); Advanced mode adds a row for reaching the machine remotely when no
machine is linked and Tailscale is not ready yet. A memory line warns that new agents may wait when free memory is low. Set up
expands a row to the commands to run yourself (the card never installs anything for you), and
Re-check re-runs the checks. The Hide (×) button dismisses the card for your browser, but it
comes back while a required check still fails. On every other view, a top banner offers the same
checklist whenever a required tool is missing.
Board
The Board (/board) shows five columns, in order: Backlog, Plan, Work, Review,
Done. Plan holds backlog issues that are being planned or already have a plan; Done shows the 12
most recent issues with a “show all” toggle. A Needs you strip sits above the columns, with item
kinds Question, Plan ready, Problems found and Stuck.
Each card shows the issue ID, a status badge, the title, labels, task progress, the agent and model,
runtime and cost, and the same action menu used everywhere else (click, or right-click). Clicking a
card opens the issue drawer; j/k move between cards and Enter opens one. The filter bar offers a
Cycle: selector (Current, All, Backlog, Canceled), an Include closed-out toggle, a force-refresh
button, and project chips when more than one project is active.
You can drag a card to reorder it inside its column. The order is not saved, and dropping a card on
another column does nothing. To move an issue forward, use its action menu instead.
Pipeline
The Pipeline (/pipeline) has a List / Board / Analytics switch (Board opens the Board view,
Analytics opens Metrics, which is experimental, so with Experimental features off it lands on Home) and five metric tiles: Active issues, Work running, Review running, Ship,
and Spend (24h). Rows are grouped into five lanes, in order: Ship, Review, Work, Plan,
Ready. Raw backlog issues are not shown; a backlog issue appears under Ready once it carries the
ready label or the tracker marks it unstarted.
Filters include phase chips (All, Ship, Review, Work, Plan, Ready), project chips, Blocked (red
checks, or a PR that cannot merge), and No PR. Filters live in the URL, so a filtered Pipeline view
can be bookmarked. A row shows priority, issue ID, title, status badge, an auto-merge badge (auto,
hold or default), project, labels, agent and model, runtime and cost; click or Enter opens the
drawer, right-click opens the action menu, and clicking the auto-merge badge switches it between auto
and hold.
Awaiting Merge
Awaiting Merge (/awaiting-merge) lists issues that are approved, have green checks and can merge,
with the empty state “Nothing awaiting merge.” At the top sit a Merge train switch (see
Merge batches) and a collapsible Merge policy summary.
Each row shows the issue ID and title, View PR, a UAT link that opens the workspace’s running
app (or Rebuild to UAT when the stack is down), an Auto/Hold toggle (see
Auto-merge), a Merge button, and an expandable panel with what to
test and the expected changes. Blocked from Merge lists PRs with failing checks or merge
conflicts, read only. A collapsed Pipeline Override section lists open PRs that are not ready,
with Force Approve and Force Merge buttons; both ask for confirmation, and both bypass review
and checks.
Issue drawer and full issue view
Clicking any issue, or opening/issues/<id> directly, opens the issue drawer over whatever view
you were on (or over Pipeline if you had none). In Simple mode, /issues/<id> opens the Simple issue
page instead. Escape or a click outside closes the drawer.
The drawer’s header has the title, the issue action strip, a link to the tracker issue, and Open
pull request. Below it, a phase strip shows Plan, Work, Review, Test, Ship, Done. Nine tabs follow:
Conversation, Tasks, Plan, Changes, Activity, Decisions, Discussion, Skills, Details. Decisions shows
only for Deft issues, and Skills hides when the issue has no project. Conversation splits into
Conversation / Terminal; Plan into Map / PRD; Changes into Files / Checks / Artifacts; Activity into
Feed / Status history.
The same content opens as a full page at /command-deck/<project>/<issue>, as a Command Deck tab
with a breadcrumb and an Open tracker link. See Command Deck.
Activity
Activity (/activity) opens on the Feed tab, with scope chips Needs you, Project (only
with a project selected) and Global, narrowed further by All, Chats and Activity
tabs. With Experimental features on, three more streams appear: Activity
Log, Detailed and TTS.
Knowledge
Knowledge (/knowledge) embeds the selected project’s OKF knowledge bundle in a read-only viewer.
With no bundle configured it shows “No knowledge bundle configured” and the command /okf init.
Edits go through /okf author; anything typed in the viewer itself is discarded. From the CLI, use
pan knowledge or pan knowledge open. See Knowledge.
Context
Context (/context) is headed “Agent context”. Pick where text applies (Machine context, Project
context, Workspace context), Edit source it, or use Preview for agent to see what a given
harness actually receives, then Save or Save & refresh outputs. From the CLI, pan context
offers list, edit, sync, diff, validate, migrate and detach. See
Context layers.
Skills
Skills (/skills) lists every skill under groups Core, Overdeck, Personal and Project, with filter
chips All, Off, Core and Overridden below, plus a search box. Turning a skill on or off here is
global, and takes effect at each agent’s next launch. From the CLI, use pan skills. See
Skills.
Settings
Settings (/settings) has a left-hand section list: Model Routing, Providers, Permissions,
Security, Computer Use, Coding Agents, Tiered Execution, Swarming, Cloister, Close-out, Merge Train,
Session Vault, Anywhere, Remote, Access Tokens, Voice, Conversations, Memory, Background AI,
Terminal, TTS, Tracker Keys, GitHub Identities, Appearance, Diff, Desktop App, Maintenance,
Telemetry, and Experimental (last). Every change saves automatically; Desktop App content only
renders inside the desktop app (see Desktop app). See Model routing,
Appearance, Issue trackers,
Auto-merge and Anywhere. See the
Settings reference for what each section controls and where it is saved.
Experimental features
The Experimental features switch lives in Settings, in the last section, Experimental (“Research-preview features that may change or be removed without notice.”). That section has three other toggles (RTK Bash compression, Claude Code Channels, and the Streamdown renderer), which this page does not document. Turning it on reveals eight views in the More menu:
With the switch off, Activity keeps its Feed but loses the Activity Log, Detailed and TTS streams.
With the switch off, a direct URL to one of the eight views above, a matching command palette entry,
or its
G shortcut lands on Home instead of the view.
Flywheel, Order books, Backlog, Board, Pipeline, Awaiting Merge, Activity (Feed), Deacon, Knowledge,
Skills, Context and Settings stay available regardless of the switch. In particular, Activity and
Skills are not experimental and stay available with the switch off, even though the toggle’s own
description in Settings currently names them.
AutoPreso
AutoPreso (/autopreso) is a voice-driven whiteboard built on Excalidraw, reached from the left nav’s More menu under Operations. It is hidden unless Experimental features is on; a direct link to /autopreso goes to Home while the switch is off.
The whiteboard has two modes. In staging, you draw on the canvas yourself. Go live sends the staged drawing to the whiteboard agent as context and makes the canvas view-only from then on; Back to staging returns to drawing it yourself. Other controls: Start mic and Stop mic, Reset session, and a warmup status (idle, warming, ready, failed).
Once live, each spoken turn is transcribed to a transcript panel and the agent edits the board in response. AutoPreso needs a microphone and browser permission, and it does not speak: there is no text-to-speech output. Speech-to-text comes from Settings → Voice → STT provider (Moonshine, a local sidecar, or Google Cloud STT with its own API key), and the whiteboard’s own model comes from Settings → Voice → AutoPreso provider and AutoPreso model. See Voice and text-to-speech.