Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.acmeagentsupply.com/llms.txt

Use this file to discover all available pages before exploring further.

​
Agent911

Agent911 is ACME’s unified control plane for agent reliability. When something goes wrong — or before it does — Agent911 gives you a single surface to understand system state, trace the cause, and follow a deterministic path to resolution.
Agent911 v0.1 is a read-only cockpit. It does not autonomously heal, repair, or restart agents. It guides operators; operators act. Never let anyone tell you otherwise.

​
The Problem Agent911 Solves

As your agent fleet grows, incidents get harder to diagnose. You have logs from five places, heartbeats from three tools, and no unified picture of what’s actually happening. When something breaks at 2am:
  • Which agent is affected?
  • What signals are available right now?
  • What should I do first?
Agent911 answers all three, immediately.

​
What Agent911 Does

Unified Telemetry Snapshot

Aggregates signals from 7 sources into a single view of system state.

Guided Recovery Playbooks

Deterministic, step-by-step operator guidance for each incident type.

Proof Bundles

Exportable evidence artifacts for post-incident review and compliance.

Incident Timeline

Reconstructed timeline of what happened, when, and in what order.

​
How It Works

Agent911 aggregates reliability signals from across the ACME stack — and from your existing infrastructure — into a coherent operator view: 
┌────────────────────────────────────────────────────┐
│                    Agent911                         │
│                                                    │
│  ┌──────────┐  ┌──────────┐  ┌──────────────────┐  │
│  │ Sentinel │  │ Watchdog │  │   OCTriage logs  │  │
│  └────┬─────┘  └────┬─────┘  └────────┬─────────┘  │
│       │             │                  │            │
│       └─────────────┴──────────────────┘            │
│                          │                          │
│                  ┌───────▼────────┐                 │
│                  │  Unified View  │                 │
│                  │  + Playbooks   │                 │
│                  └────────────────┘                 │
└────────────────────────────────────────────────────┘

​
What Agent911 Does NOT Do

This list is intentional. Trust requires precision.
In v0.1, Agent911 does not:
  • Automatically restart or repair agents
  • Modify configurations without explicit operator action
  • Make autonomous decisions about recovery
  • Replace the need for an operator on-call
It is a guidance and visibility tool. Every action taken during an incident is operator-initiated.

​
Telemetry Sources

Agent911 v0.1 aggregates from 7 signal sources:
SourceSignal Type
SentinelRuntime anomalies, stall detection
WatchdogHeartbeat / liveness probe results
OCTriageIncident classification, evidence bundles
DriftGuardSession-to-session behavioral delta
RadCheckBaseline reliability score history
LazarusRecovery readiness posture
Native logsRaw process/gateway logs (when available)

​
Snapshot: What You See

When you open Agent911, you see a snapshot — a coherent view of system state at a point in time:
  • Current health signals across all agents
  • Recent anomalies and their correlations
  • Routing/governance status (if SphinxGate is configured)
  • Recovery readiness posture (if Lazarus is configured)
A snapshot doesn’t just show raw signals — it contextualizes them. If Sentinel detected a stall 4 minutes ago and Watchdog shows a missed heartbeat at the same time, Agent911 surfaces these as a correlated event, not two separate alerts.
→ See the full Snapshot reference

​
Guided Playbooks

For each incident type Agent911 recognizes, it provides a deterministic recovery playbook — a numbered list of operator actions in recommended order: 
Incident: STALL (High Confidence: 91%)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Playbook: STALL_RECOVERY_v2

Step 1: Verify Sentinel alert context (attached)
Step 2: Check external API call status (OpenAI timeout suspected)
Step 3: Run `acme radcheck` to confirm no secondary issues
Step 4: Review Lazarus readiness before restart
Step 5: If ready, restart agent with `acme agent restart <name>`
Step 6: Confirm Watchdog liveness post-restart
Playbooks are deterministic — same incident type, same playbook — which means any team member can run them, not just the person who knows the system by heart.

​
FindMyAgent (Included)

Agent911 includes FindMyAgent — live presence and operator awareness for your agent fleet. See which agents are active, stalled, or haven’t checked in.

​
CLI Reference

# Open Agent911 dashboard
acme agent911

# Get current snapshot (JSON)
acme agent911 snapshot --format json

# Get snapshot for a specific agent
acme agent911 snapshot --agent <agent-name>

# View active incidents
acme agent911 incidents

# Export proof bundle
acme agent911 bundle --incident <incident-id> --output bundle.tar.gz

# View playbook for incident type
acme agent911 playbook STALL

​
Pricing

Agent911 is a paid expansion product. Most teams add it after establishing Sentinel — typically when fleet size grows or when incidents start recurring. See Pricing for current rates.

​
Next Steps

Snapshot Explained

Deep dive into what a snapshot contains and how to read it.

FindMyAgent

Agent presence and operator awareness, included with Agent911.

Sentinel

The detection layer that feeds Agent911. Start here first.

Lazarus

Verify recovery readiness before you need it.