AgentLattice/Docs
Get Started
Navigate docs

Releases

Dashboard

The AgentLattice dashboard at agentlattice.io is your central command for AI agent governance. It shows you the health of your agent fleet at a glance, surfaces actions that need your attention, and provides the audit evidence your compliance team needs.

Fleet Health Overview

The top of the dashboard shows a single, clear signal: is your fleet healthy?

  • Healthy (green) — No active incidents, no agents halted or killed, no stale approvals. Your fleet is operating normally.
  • Attention (yellow) — One or two active incidents, agents in warning or throttle state, or approvals waiting longer than 24 hours. Review recommended.
  • Critical (red) — Three or more active incidents, any agent halted or killed, or a critical-severity incident. Investigate immediately.

Below the status indicator, a narrative summary tells you exactly what is happening in plain language: how many agents are active, how many actions were processed today, and what (if anything) needs your attention. No dashboards full of numbers to interpret. One sentence, one status.

A 7-day sparkline next to the status indicator shows the anomaly event trend over the past week. The sparkline color matches the fleet status, and a trend label indicates whether anomalies are rising, falling, or stable. This gives you an instant visual sense of whether things are getting better or worse without clicking into the detail pages.

Live Updates

The dashboard updates automatically when your fleet data changes. There is no need to manually refresh the page.

When an incident is created, an approval comes in, an anomaly is detected, a policy changes, or an enforcement event fires, the dashboard refreshes within 1-2 seconds to show the latest data. This works whether you are actively watching the dashboard or returning after stepping away. If you close your laptop and reopen it, the dashboard catches up on any changes that happened while you were away.

No configuration is required. Live updates are enabled for all dashboard users.

Copy Summary

A clipboard button next to the persona toggle lets you copy the current fleet status as formatted text. This is useful for pasting into Slack, email, or incident reports. The copied text includes fleet status, agent count, incident count, pending approvals, and chain integrity status.

Since Your Last Visit

When you return to the dashboard after being away, a brief summary appears below the greeting showing what changed: new incidents, new approvals, or both. This helps you quickly orient to what happened since you last checked. The summary disappears once there is nothing new to report.

Persona Views

Different roles need different information first. The dashboard adapts to your perspective with three persona views:

  • CISO — Leads with fleet health, security posture, and compliance status. Incidents, anomalies, delegation chains, policy coverage, and audit chain integrity are front and center.
  • VP Engineering — Leads with fleet health, pending actions, and agent fleet status. SDK usage and audit summary follow. Security and compliance details are available but not the default focus.
  • Developer — Leads with fleet health, pending actions, and agent status. SDK integration details are prominent. Security and compliance sections are hidden to reduce noise.

Your persona preference is saved locally and persists across sessions. Switch anytime using the toggle at the top of the dashboard.

Actions Required

The actions section surfaces everything that needs a human decision, sorted by urgency:

  1. Critical incidents appear first — active security events that need immediate investigation
  2. Pending approvals come next, oldest first — agent actions waiting for human review, with time remaining before timeout
  3. Warning-state agents follow — agents that have triggered behavioral anomaly detection

Each item links directly to the relevant detail page so you can act without searching. When there are no pending actions, the section shows "All clear."

Agent Fleet

A compact view of your registered agents and their current status. Each row shows:

  • Agent name — The label you assigned when creating the agent
  • Status — The agent's current circuit breaker state: Monitoring (normal), Calibrating (learning baseline), Warn, Throttle, Halt, or Killed
  • Risk score — A 0-100 score combining the agent's circuit breaker state with recent anomaly count
  • Last active — When the agent last took an action

The full agent management view is one click away for deeper investigation, key rotation, or creating new agents.

Security Posture

The security posture section (visible in CISO view) shows three key metrics:

  • Active incidents — How many security incidents are currently open, with severity breakdown
  • Anomalies (24h) — How many behavioral anomalies were detected across all agents in the last 24 hours
  • Delegation chains — How many active delegation chains exist, how many were granted, and how many were revoked

High counts are color-highlighted so patterns are visible at a glance. Click through to the full security activity page for investigation.

Summary Strip

At the bottom of the dashboard, three summary segments provide quick reference:

  • Audit — Total actions logged today, chain verification status, and a link to the full audit log
  • SDK / MCP — Active API key count, total API calls, and a link to key management
  • Compliance (CISO view) — Active policy count, chain signing status, and a link to the policy editor

Approval Queue

The approval queue shows every agent action currently waiting for human review. An action enters the queue when its governing policy has require_approval set and the action's conditions pass.

Each pending approval displays:

  • Agent name: Which agent is requesting approval
  • Action type: What the agent wants to do
  • Metadata: Context the agent provided — PR number, repository name, migration script, environment, sensitivity level, or any other data the agent included in the execute() or gate() call
  • Data accessed: What resources the agent reports accessing, with type and sensitivity classification
  • Time remaining: How long until the approval window expires (set by the policy's timeout)
  • Policy name: Which policy requires this approval

From the queue, you can act directly from each tile:

  • Approve once — unblocks this specific request. Future actions of the same type still require approval.
  • Approve forever — approves this request and disables the approval requirement on the matching policy, so future actions of this type auto-execute. Requires admin role. A confirmation step prevents accidental policy changes.
  • Deny — denies the request with a confirmation step. The agent receives a structured denial.

Members and admins can approve or deny. Viewer-role accounts have read-only access to the queue.

Approval decisions are recorded in the audit trail with the reviewer's identity and timestamp. This creates the segregation-of-duties evidence that compliance frameworks require.

If an approval times out, the action is automatically marked as timed out and the agent is notified. Timed-out actions appear in the activity feed so you can identify patterns.

Settings

The settings area covers workspace-wide configuration:

API Keys

  • View and manage API keys for all agents in your workspace
  • Rotate keys when needed — the old key is immediately invalidated
  • Each agent has its own key, so revoking one agent's access does not affect others

Workspace Configuration

  • Workspace name: How your workspace appears in the dashboard and audit exports
  • Team members: Invite colleagues to your workspace. Members can view the activity feed, manage agents, edit policies, and review approvals based on their role
  • Auto-discover action types: Toggle automatic cataloguing of action types that agents call via gate(). Enabled by default for new workspaces. When on, discovered types appear on the Policies page so you can identify coverage gaps
  • Domain auto-join: Configure email domains that allow automatic membership
  • Invitation management: Send invitations to specific email addresses, view pending invitations, and manage roles

Getting Started

When you first log in after signing up, the dashboard detects that your workspace is new and shows a guided setup experience:

  1. Register your first agent — Create an agent identity with a dedicated API key
  2. Set up a governance policy — Define your first rule for what agents can do
  3. Connect your agent — The wizard asks how you will integrate:
    • Write gate() manually — direct SDK integration for new agents you are building
    • al-instrument — one command adds governance to an existing Python or LangChain agent
    • al-instrument-ts — one command for existing Claude SDK, Vercel AI, or OpenAI TypeScript agents
    • /instrument skill — Claude Code auto-instruments any framework, including CrewAI, AutoGen, and custom patterns

Each step links directly to the relevant page. As you complete steps, they are marked done. Once all three are complete, the full dashboard appears with your live fleet data.

This zero-to-first-event path confirms that your account, agent, policy, and audit trail are all working end to end.

The sidebar provides access to all major sections of AgentLattice:

  • Dashboard — Fleet health overview and live event feed
  • Approvals — Pending agent actions waiting for human review
  • Agents — Agent identities, delegations, and governance policies
  • Activity — Audit log, security operations center, and incident timeline
  • Settings — Team members, API keys, principals, and security configuration

At the bottom of the sidebar, two utility links provide quick access to help and documentation:

  • Help opens a context-sensitive help panel on the right side of the screen. The panel automatically shows guidance relevant to your current page. For example, when you are on the Approvals page, the help panel explains approval workflows, timeouts, and segregation of duties.
  • Docs opens the full AgentLattice documentation site in a new tab.

User Menu

Click the three-dot menu next to your email at the bottom of the sidebar to access:

  • Appearance — Switch between Light, Dark, and System (follows your OS preference) themes. Your choice is saved and persists across sessions.
  • Sign out — Ends your session and returns to the login page.