Skip to main content
The web dashboard is your visual control center for everything Comis does. From any browser, you can chat with your agents, check channel connections, review memory, tune configuration, and monitor system health — all without touching the command line. Who it’s for: operators, developers, and team owners who would rather click than YAML. Under the hood: the dashboard is a single-page app built with Lit (web components), Vite, and Tailwind, served by the Comis gateway. It talks to the daemon over JSON-RPC and subscribes to a Server-Sent Events (SSE) stream for real-time updates. There is no separate React/React Flow framework — pipeline canvases, charts, and trees are all custom Lit components.
The Comis web dashboard uses a dark theme — there’s no light-mode toggle, so it’ll look its best on a dark desktop or browser profile.

Getting Started

1

Open the dashboard

Navigate to http://localhost:4766 in your browser. This is the default address where the gateway serves the dashboard. (Override with gateway.host and gateway.port in ~/.comis/config.yaml.)
2

Enter your gateway token

On the login screen, paste the gateway token from your configuration file. You can find it in ~/.comis/config.yaml under the gateway.tokens section, or generate a new one in Security > API Tokens after first login.
3

Click Connect

After entering your token, click Connect. The dashboard validates the token via system.ping, then loads the main dashboard.
Your token is stored in sessionStorage under the key comis_token. That means it is automatically cleared when you close the browser tab, and never persists to disk. The next time you open the dashboard, you will need to paste the token again. This is a deliberate security measure to prevent stale sessions on shared machines and to limit the blast radius of an XSS bug. There is no refresh token — if your session expires (gateway restart, token revoke), you re-enter the token.

Real-time updates

The dashboard listens to an SSE stream at /sse/subscriptions/<id> for events the daemon emits as work happens. The most user-facing events are:
EventWhat you see
approval:requestedThe Approvals badge in the sidebar increments. A toast appears.
approval:resolvedThe badge decrements; the Pending Approvals queue updates.
system:errorA red toast describing the failure (e.g., a channel disconnected).
metrics:updateStat cards on the Dashboard and Observability views refresh.
delivery:statusThe Delivery view’s trace table updates as messages move through the queue.
graph:node:status, graph:progressThe Pipeline Monitor turns nodes green/red as they execute.
subagent:spawned, subagent:statusThe Sub-Agents view’s run list refreshes.
Views that are not SSE-driven (Memory, Sessions, Scheduler) refresh on demand or via a 30-60 second polling controller. Each view’s page documents its update strategy.

What You See

Once connected, the dashboard is organized into three main areas: a sidebar on the left, a top bar across the top, and a content area that fills the rest of the screen. The sidebar is your primary way to navigate between different views. It contains 23 items organized into four sections — Home (1), Operate (7), Observe (6), and Configure (9) — plus a Setup item and Logout button below a divider. Each section header is a non-clickable visual label that groups related items together.
The sidebar shows 23 items, but the underlying SPA actually serves a few more URLs:
  • Agent detail / editor / workspace are reachable from the Agents view (clicking an agent row).
  • Channel detail opens from a Channels card.
  • Session detail opens from a row in Sessions.
  • Pipeline builder, monitor, history, history detail are sub-views of Pipelines.
  • Approvals still has its own sidebar entry but the page now redirects to the Security > Pending Approvals tab (the canonical home for approvals).
In total, 38 URL patterns route to 23 distinct view components. This page documents every component you can reach.

Home

ItemWhat It Does
DashboardAt-a-glance overview with agent cards, channel badges, and activity feed

Operate

ItemBadgeWhat It Does
AgentsAgent countView, configure, suspend, and delete your AI agents
ChannelsChannel countMonitor connection status for each messaging platform
MessagesBrowse and filter all messages across channels with type-based views
ChatTalk to your agents directly through the browser
SessionsSession countView conversation histories and session statistics
Sub-AgentsMonitor spawned sub-agent tasks, outputs, and lineage
PipelinesBuild and monitor multi-agent execution graphs visually

Observe

ItemBadgeWhat It Does
OverviewError countSystem metrics summary with stat cards and 24-hour activity chart
Context EnginePipeline metrics, layer waterfall, and DAG visualization
Context DAGInteractive DAG browser for inspecting context node graphs
BillingToken usage and cost breakdown by provider and agent
DeliveryEnd-to-end message delivery traces and success rates
DiagnosticsSystem diagnostic event log with category filtering

Configure

ItemBadgeWhat It Does
SkillsBrowse installed skills per agent with an agent selector dropdown and scope toggle
MCP ServersManage MCP server connections, view tool inventories, and monitor health
ModelsSee which AI models are available and their configurations
MemorySearch and browse your agents’ stored memories
SchedulerManage cron jobs, view run history, and monitor heartbeats
SecurityAudit log, token management, approval rules, and security events
MediaTest media processing (vision, TTS, STT, image generation) and configure providers
ApprovalsPending approvalsReview and act on pending approval requests
ConfigEdit your Comis configuration in real time
Below a divider, you will also find Setup — the guided setup wizard for first-time configuration. In the sidebar footer, the Logout button ends the session and returns to the login screen. Badge counts update in real time via Server-Sent Events (SSE). When agents come online, channels connect, sessions start, errors occur, or approvals arrive, the corresponding badge updates automatically without any manual refresh. The sidebar is collapsible. Click the collapse button at the bottom to shrink it to icons only, giving more room to the content area. Your preference is remembered across sessions. On mobile devices, the sidebar becomes an overlay drawer that opens when you tap the hamburger menu icon.

Top Bar

The top bar sits across the top of the screen and shows the Comis logo on the left. On the right, you will see:
  • Connection status — a colored dot with icon and text label that tells you whether the dashboard is connected to your Comis daemon. The status uses both color and iconography for WCAG AA accessibility:
    • Green (check-circle icon) = Connected and working normally
    • Yellow (refresh-cw icon, pulsing) = Reconnecting — the dashboard is trying to restore the connection
    • Red (x-circle icon) = Disconnected — check that your Comis daemon is running
  • Notification bell — shows a badge count when there are pending items
  • User menu — click the avatar circle to see your token identifier and a Logout option
If you see a red dot, the most common fix is to make sure the daemon is running. Open a terminal and run comis daemon status to check.

Content Area

The main content area switches based on which item you select in the sidebar. Each view is documented in detail on its own page.
The Skills view lets you browse installed skills per agent with a scope toggle (agent vs shared). It shows built-in tool descriptions, categories, and parameter hints. The Models view shows all configured AI models with their provider, status, context window, and pricing, plus model aliases and a connectivity test. The Setup wizard walks through initial configuration including provider API keys, channel setup, and first agent creation. For in-depth configuration, see Skills, Agents - Models, and Installation.

Dashboard

The Dashboard is the landing page after login (/dashboard route). It provides an at-a-glance overview of your entire Comis system, organized into several rows:
  • Stat cards — six clickable cards showing Active Agents (ratio of active to total), Sessions, Messages Today, Tokens Today, Cost Today (USD), and Errors. Each card links to its detail view. Messages, Tokens, and Cost cards show trend arrows comparing to previous values.
  • System Health card — displays daemon uptime, memory (RSS), event-loop delay, Node.js version, MCP server count, and an optional CPU progress bar. Links to the Observability Overview.
  • Context Engine Summary card — shows prompt cache hit rate, budget utilization percentage, total evictions, and re-reads count. Links to the Context Engine view.
  • Sparkline charts — two mini charts showing Token Usage (24h) and Cost Trend (7d) with clickable navigation to the Billing view.
  • Agents section — a grid of agent cards, each showing the agent’s name, provider, model, status indicator, messages today, tokens today, and per-agent cost badge.
  • Channels section — channel badges showing each platform’s type, name, connection status, enabled state, and uptime.
  • Activity feed — a live-updating feed of recent system activity, powered by Server-Sent Events (SSE).
All stat cards update in real time via SSE events. RPC-sourced data (sessions, messages, tokens, cost, errors) refreshes every 60 seconds. The dashboard shows loading skeletons during initial data fetch and a retry button if an error occurs.

Keyboard Shortcuts

The dashboard supports keyboard shortcuts for fast navigation and common actions. Press ? at any time to see the full shortcuts help panel.
ShortcutActionWorks in Text Fields
Ctrl+K / Cmd+KOpen command paletteYes
EscapeClose current overlay (command palette, shortcuts help, mobile sidebar)Yes
?Toggle keyboard shortcuts help panelNo
G then DGo to DashboardNo
G then AGo to AgentsNo
G then CGo to ChatNo
G then SGo to SessionsNo
G then OGo to ObservabilityNo
Two-key shortcuts (the G+letter sequences) have a 500ms timeout. Press G, then press the second key within 500 milliseconds. If you wait too long, the sequence resets. Most shortcuts are disabled when your cursor is in a text field — only Ctrl+K / Cmd+K and Escape work inside inputs.

Command Palette

Press Ctrl+K (or Cmd+K on macOS) to open the command palette — a quick-access search overlay that lets you jump to any view, agent, session, or command without using the sidebar. The palette searches across four categories:
CategorySourceExamples
Views24 navigable views from the sidebar and sub-viewsDashboard, Agents, Messages, Billing, Context Engine
AgentsDynamic list from pollingdefault, research-agent, support-bot
SessionsDynamic list from pollingtelegram:user123:chat456
CommandsBuilt-in actionsRefresh Data, Toggle Sidebar, Logout, Show Keyboard Shortcuts
Use the arrow keys to navigate results, press Enter to select, and press Escape to close. The search is fuzzy — typing partial names or abbreviations will find matches. Each card below links to its own page. The 23 distinct views are grouped by sidebar section.

Operate

Chat Console

Talk to your agents in real time through the browser.

Agents View

Browse, configure, suspend, and delete agents (list + detail + editor).

Workspace Manager

Browse and edit each agent’s workspace files; commit and roll back via git.

Channels View

Status, restart, configure each messaging platform connection (list + detail).

Message Center

Send, react, edit, delete messages and run platform actions.

Sessions View

View conversation histories, transcripts, context state, and metrics.

Sub-Agents

Monitor spawned sub-agent runs, costs, and parent-child lineage.

Pipelines

Build and monitor multi-agent execution graphs (builder + monitor + history).

Observe

Observability Overview

Stat cards and 24-hour activity chart — the entry point to all observability tabs.

Context Engine

Pipeline metrics, layer waterfall, and cache hit rate.

Context DAG Browser

Inspect the DAG of context nodes and summaries.

Billing

Token usage and cost breakdown by provider and agent.

Delivery

End-to-end message delivery traces and success rates.

Diagnostics

System diagnostic event log with category filtering.

Configure

Skills

Browse installed prompt skills and built-in tools per agent.

MCP Servers

Manage MCP server connections, view tool inventories, and monitor health.

Models

Configure providers, model catalog, aliases, and defaults.

Memory View

Search, browse, and delete memory entries with type/trust filters.

Scheduler View

Manage cron jobs, run history, and heartbeat monitoring.

Security View

Audit log, tokens, approval rules, secrets, and approval queue.

Media

Test STT/TTS/Vision/Document/Video/Link processing and configure providers.

Config Editor

Edit your Comis configuration — changes save to disk and survive restart.

Setup Wizard

Re-run the guided onboarding flow for first-time configuration.