Dashboard

Dashboard is Agent Network's web management interface, providing real-time monitoring and task management capabilities.

Current Dashboard

Start Mode	Tech Stack	Default URL	Notes
anet hub dashboard	Next.js 16	http://localhost:3000	CLI starts @sleep2agi/agent-network-dashboard@${tag} via npx; tag comes from dashboardReleaseTag() (cli.ts): defaults to @preview, overridable via the ANET_DASHBOARD_VERSION env var (no hardcoded version pin — see Preview channel below). Thin cookie-proxy mode (no service token)
Standalone deploy	Next.js 16	Custom	Configure it with the CommHub URL

TIP

anet hub start starts only CommHub Server. Start the Web UI in another terminal with anet hub dashboard.

v0.10.0 / dashboard 0.5.0 new — Hero 3 network/node front-end 8/8 surfaces ✅

Shipped with v0.10.0 release Phase 2: all 8 surfaces complete (§3.A prefix-group fix / §3.B sweep retire / §3.C recent-panel hide / §3.D grid default view / §3.E hover detail card / §3.F server-health ring tint wired to #99 endpoint / §3.G fullscreen mode / §3.I canvas brand mark; §3.H dropped per RFC Q2 review) plus 19+ rounds of typography + corner-radius cascade polish. The server-health ring tint reads from GET /api/server/:host/health; the agent hover card renders per-agent process_telemetry (rss / cpu_pct / uptime_seconds / in_flight_count, #142 shipped in agent-node@2.4.0 + server schema aligned in commhub-server@0.8.2).

§3.E / §3.F data sources require agent-network ≥ 2.2.1 on the default anet hub start path (v0.10.1 hotfix bumped PINNED_SERVER_VERSION from 0.8.0 to 0.8.2). Older versions still launch commhub-server@0.8.0, where the #99 endpoint does not exist — the ring-tint data source fails and the hover card's process_telemetry is all null.

v0.10.2 Hero D — dashboard 0.5.1 topology prefix label Option C + disk render

Shipped alongside the v0.10.2 release:

• Hero D topology node prefix labels — Option C implementation (#147 acked 5/16 + Option C design pass) — node → group edge label distinguishability finally lands
• Disk telemetry hover-card rendering (disk_total_gb / disk_used_gb / disk_avail_gb wired to the GET /api/server/:host/health response; source is agent-node ≥ 2.4.1 host-telemetry's df -k sampling; older agents and Windows render — rather than a misleading 0)
• 100+ rounds of typography + corner-radius cascade polish

v0.10.8 dashboard 0.5.3 — Servers panel UI copy fix + TopoGraph density tier

Shipped alongside the v0.10.8 release (#157 Fix #1, caught by Vincent):

• Servers panel "not reported" copy made accurate: the early placeholder strings agent rollup pending hub ≥ 0.8.2-preview / disk metric pending hub ≥ 0.8.2-preview were buried during the 0.8.2 upgrade window. After every hub already became ≥ 0.8.2, they still rendered (misleading users into thinking their hub version was too old). Now agent rollup not reported by hub / disk metric not reported by hub — accurately conveying "this hub didn't report it right now".
• data-server-{agents,disk}-missing="true" Playwright hooks: new since dashboard 0.5.3, available for e2e validation of hub-side telemetry coverage.
• TopoGraph density-tier polish (purely additive, no UX change): the canvas state attribute data-topo-fleet-density-tier ∈ {empty, sparse, normal, dense, very-dense} exposes a 12th observable testing surface. Tier boundaries (sparse 1-3 / normal 4-15 / dense 16-30 / very-dense 31+) line up with the dense-layout collapse gate.

Root causes #2 + #3 located, deferred

• #2 (v0.10.9 candidate): missing dedupe when one hostname appears multiple times can double-count servers
• #3 (v0.11.0 candidate): status=offline vs telemetry mismatch (telemetry still reports but the SSE last_seen has timed out) — needs system-level status reconciliation

::: :::

Page Overview

The current primary navigation is Nodes / Overview / Schedules / SkillHub / Tasks / Servers / Providers / Admin / Settings (Admin is visible only to system-level admins). Low-frequency routes such as Messages remain available through a direct URL or the command palette. ChatPanel is embedded and opens from an online agent card in Overview or a node view in Nodes; there is no standalone Chat navigation page.

Overview

The overview page displays the overall network state:

• Online agent count: Currently connected agent nodes
• Task statistics: Pending / In progress / Completed / Failed
• Network activity: Message volume trends over the last 24 hours
• Topology graph: Communication relationship visualization between agents

Tasks

The task management page displays the full lifecycle of all tasks:

Column	Description
Task ID	Unique identifier (clickable for details)
From	Sender alias
To	Recipient alias
Priority	Priority level (high / normal / low)
Status	Status (created / delivered / acked / running / replied / failed / cancelled / expired — 8 states total; verified at server/src/db.ts:94 status TEXT NOT NULL DEFAULT 'created'; cancel_task covers 4 cancellable states including created; full state machine: Task lifecycle)
Content	Task content preview
Created	Creation time
Duration	Time from creation to completion

Action buttons:

• Send Task -- Select target agent + enter content + set priority
• Retry -- Re-deliver failed/cancelled tasks
• Cancel -- Cancel pending tasks
• Reassign -- Transfer a task to another agent

Status filters:

[All] [Pending] [In Progress] [Completed] [Failed] [Cancelled] [Expired]

Task detail modal:

Task ID:    t_a1b2c3d4
From:       commander
To:         coder-1
Priority:   normal
Status:     replied
Content:    Write a Hello World Python script
Result:     ```python\nprint("Hello World")\n```
Created:    2026-04-12 10:00:00
Delivered:  2026-04-12 10:00:01
Started:    2026-04-12 10:00:03
Completed:  2026-04-12 10:00:15
Duration:   15s

Event Log:
  10:00:01  delivered → coder-1
  10:00:03  acked by coder-1
  10:00:03  running
  10:00:15  replied by coder-1

Nodes

The node management page displays detailed information about all agent nodes:

Column	Description
Alias	Agent name
Status	State (idle / working / offline / error)
Runtime	Runtime engine (claude-agent-sdk / codex-sdk / claude-code-cli)
Model	Model name
Server	Host server
Last Seen	Last heartbeat time
Task	Currently executing task

Status indicators:

Color	Status	Meaning
Green	idle	Online, waiting for tasks
Yellow	working	Processing a task
Red	error	Runtime error
Gray	offline	Offline

Messages

Real-time message stream showing all inter-agent communication:

15:00:42  commander → coder-1: [task] Write a sorting algorithm
15:00:43  [SSE] coder-1 received push
15:00:45  coder-1 → commander: [reply] Done, implemented with quicksort
15:01:05  commander → all: [broadcast] Take a 5-minute break

Message type labels:

Label	Meaning
[task]	Formal task
[reply]	Task reply
[message]	Chat message
[broadcast]	Broadcast
[ack]	Acknowledgement

Message data comes from CommHub REST APIs. Agents receive push events through /events/:alias SSE connections and write state back to the Hub.

ChatPanel

ChatPanel lets you talk to agents directly in the browser. It is not a standalone page: open it from an online agent card in Overview or from a node view in Nodes:

1. Select a target agent (from the online list)
2. Enter your message
3. Choose the send type:
   • Task -- Formal task, the agent will process and reply
   • Message -- Chat message, the agent won't auto-process
4. View the agent's reply

Admin

Only system-level admins

The Admin panel is visible to users with users.role='admin' — that's a system-level role (granted automatically to the first registered user), not a network_members.role='admin' (per-network role). A network-level admin can only see their own /api/audit-log rows like any other role; reading all rows requires the system-level admin.

Admin features include:

• User Management -- View all registered users (/api/users — system-level admin only); role changes currently go through REST PUT /api/networks/:id/members/:user_id (owner only — see API — PUT members). CLI has no promote / demote sub-command yet (queued for v0.11+ / unscheduled).
• Network Management -- View all networks, members (plan-quota is partially enforced in v0.8: createNetwork still enforces max_networks_owned; other quota items are dormant — see networks — quota limits)
• System Statistics -- Server load, database size, connection count
• Audit Log -- Detailed records of all operations (/api/audit-log endpoint + Dashboard Audit Log page; system-level admin sees everything, other roles only see their own rows)

Audit log example (actual 19 actions):

Time	User	Action	Details
10:00:01	alice	register	username=alice
10:00:05	alice	password_changed	(via anet passwd)
10:00:10	alice	network_renamed	dev → development
10:00:15	alice	member_added	u_bob_xxx as member

The older example listed create_network as an audit action — it does not exist. security.md audit log already documents this: POST /api/networks does not call logAudit, so the audit_log table never contains create_network or network_created rows. The actual 19 actions are (RFC-010 node-rename added 3 node_rename_* actions; 18 go through the logAudit() helper + 1 (password_reset_by_admin) goes via direct INSERT at auth.ts resetUserPassword()): register / login / login_failed / login_rate_limited / password_changed / password_reset_by_admin / network_renamed / network_deleted / network_joined / member_added / member_role_changed / member_removed / token_created / token_revoked / node_token_created / node_rename_prepared / node_rename_committed / node_rename_aborted / invite_created.

Settings

The settings page manages personal configuration:

• Profile -- Edit display name, email
• Password -- Change login password (current stable Dashboard 0.6.0 UI not yet shipped; use CLI anet passwd — see account-system / Change Password)
• Token Management -- Create / view / revoke API tokens
• Network Settings -- Current network config (owner/admin only)
  • Rename network
  • Create invite codes
  • Manage member roles (current stable Dashboard 0.6.0 partial; CLI has anet network invite but no promote / demote sub-commands — role changes currently go through REST PUT /api/networks/:id/members/:user_id. CLI sub-commands queued for v0.11+ / unscheduled.)
  • Delete network

Token management interface:

Name	Scope	Network	Last Used
user-login	user	-	2026-04-12 10:00
node:coder-1	network	default	2026-04-12 09:55
dashboard	full	default	2026-04-12 10:01

Actions: [+ Create Token] [Revoke]

Access

Local Dashboard

# Terminal 1: start Server
anet hub start --port 9200

# Terminal 2: start Dashboard
anet hub dashboard

# Optional: explicitly use the Dashboard already installed on PATH
# (the CLI never auto-terminates this process)
ANET_DASHBOARD_LOCAL=1 anet hub dashboard

# Open in browser
open http://localhost:3000

Standalone Dashboard

# Start with Docker Compose
docker compose up dashboard

# Or deploy to Vercel
cd agent-network-dashboard
vercel deploy --prebuilt --prod

The standalone Dashboard requires the following environment variables:

Variable	Description
COMMHUB_URL	CommHub Server address
COMMHUB_AUTH_TOKEN	Legacy global auth token; soft-deprecated in v0.8, removed in v1.0. v0.4.2 Dashboard runs as a thin cookie proxy and no longer needs it.
COOKIE_INSECURE	Set to 1 for dev mode (HTTP)

Real-Time Update Mechanism

The Dashboard keeps data current through three data surfaces:

1. REST queries: Reads /api/status, /api/tasks, /api/messages, and related endpoints
2. Dashboard's own SSE: The Dashboard subscribes to the /events/<username> user channel using the logged-in username, receiving server-pushed events directly (e.g. RFC-010's node.renamed, the #84 SSE channel fix — see REST API SSE endpoint)
3. Agent SSE: Agents subscribe to /events/<alias> with their own node alias; when tasks arrive, agents update Hub state that the Dashboard reads

Performance Note

If you have more than 50 agents, consider using the standalone Dashboard and disabling real-time message streaming in favor of manual refresh.

Preview channel (next stable WIP)

@sleep2agi/agent-network-dashboard@preview carries the next-gen UI under active polish. Current preview pin auto-syncs with the CLI preview tag @sleep2agi/agent-network@preview; the actual version is whatever's on the npm preview tag (frequent iteration, this doc doesn't pin a specific number).

New capabilities (vs the current stable — the stable dashboard also keeps iterating; if any item below has already landed in stable, the npm package page's dist-tags are authoritative):

• Cmd / Ctrl + K command palette: keyboard-driven navigation, search, command invocation
• ? keyboard shortcut overlay: all hotkeys at a glance
• Global health banner: red / amber / green tri-color + CTA + dismiss
• KPI card hover popover: working / idle / offline breakdown
• EmptyState — 7 variants: tailored illustrations for post-login, empty network, zero tasks, etc.
• Topology redo + 38+ rounds of polish (#112 + #116):
  • Dual layouts: grid (rows-by-group) + ring (orbiting the central hub); bottom-right G / R toggle
  • Node interactions: mount fade-in (nodes phase in on join) + hover ring focus (halo on hover) + click ripple + label scale on zoom (counter-scales so text stays readable)
  • Edge tiers: arrow tier (message frequency drives arrow thickness / opacity) + offline dim (edges to offline nodes fade)
  • Grouping: group-box hover (hovering a group title highlights every member) + dashed separators between groups
  • Side controls: minimap (bottom-left thumbnail with draggable viewport box) + cwd tooltip (hover a node to see its project_dir) + S/M/L size toggle
  • Light mode: 24px pulse on the central hub (fixes stable's "invisible on light mode" P0)
• ServersDrawer (#119): a right-hand drawer that aggregates agents by physical machine (hostname + ip) with live CPU load / memory bars and an agent count. Backed by GET /api/servers (10-min stale-mark side effect + bare JSON array response, index.ts GET /api/servers). Multiple agents on the same host collapse into one row; older agents without host telemetry are bucketed under an unknown host group.
• Tasks status tabs: color-coded dots + mobile horizontal scroll
• Mobile audit fixes: banner yields to hamburger / UserBar iconified
• Sidebar "Quick search ⌘K" chip: mobile launcher entry
• LoadingSkeleton redo: mirrors Overview layout + brand-pulse rhythm
• P0 light-mode contrast sweep: text-{color}-300/400 → -700

Try it:

# Upgrade CLI to preview
npm i -g @sleep2agi/agent-network@preview
anet -v                                    # should show 2.3.0-preview.N (current preview channel; latest stable is 2.2.x)
anet hub dashboard                          # npx auto-pulls the current preview version

The default mode still starts the channel-matched release through npx. The CLI only replaces a stale npx Dashboard that it previously recorded and whose PID, birth fingerprint, and command all match exactly. It refuses automatic termination when the port belongs to an unknown process, inspection is incomplete, or a global install is in use. It never uses pkill, killall, or process-name prefix matching.

To reuse a global installation from PATH, explicitly set ANET_DASHBOARD_LOCAL=1. That Dashboard remains operator-managed; rerunning the command does not kill or replace it.

Or bypass the CLI entirely:

npx -y @sleep2agi/agent-network-dashboard@preview --ip 0.0.0.0

Preview is not backward-compatible

The preview channel iterates continuously and is not auto-promoted to latest. Stick with stable (@sleep2agi/agent-network@latest) for production.