Platform Overview
ACME is a reliability stack for AI agent operators. The products aren’t a bundle of unrelated tools — they form a coherent system designed around how agent failures actually happen and how operators actually respond to them.
The Reliability Loop
Agent reliability follows a loop. ACME tools map to each phase:
┌──────────────────────────────────────────────────────────────┐
│ │
│ ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌───────┐ │
│ │ DETECT │───▶│ TRIAGE │───▶│ RECOVER │───▶│VERIFY │ │
│ └──────────┘ └──────────┘ └──────────┘ └───────┘ │
│ │ │ │ │ │
│ RadCheck OCTriage Agent911 Lazarus │
│ Sentinel DriftGuard Watchdog │
│ Watchdog SphinxGate │
│ FindMyAgent │
│ │
└──────────────────────────────────────────────────────────────┘
▲ │
└───────────────────────────────┘
Transmission (signal bus)
Layer by Layer
Detect
Tools that identify problems — before users do.
| Tool | What It Detects |
|---|
| RadCheck | Point-in-time reliability risk (0–100 score) |
| Sentinel | Real-time stalls, silence gaps, runtime anomalies |
| Watchdog | Missed heartbeats, liveness failures, throughput collapse |
| DriftGuard | Long-horizon behavioral drift across sessions |
| FindMyAgent | Fleet presence — who’s up, who’s stalled, who needs attention |
| SphinxGate | Unauthorized routing, policy violations, audit anomalies |
Triage
Tools that help you understand what happened and what to do.
| Tool | What It Classifies |
|---|
| OCTriage | Incident type, root cause, evidence bundle, next steps |
| DriftGuard | Drift patterns and memory integrity issues |
Recover
Tools that coordinate and guide recovery.
| Tool | What It Provides |
|---|
| Agent911 | Unified control plane — telemetry, playbooks, proof bundles |
| Watchdog | Escalation and handoff to recovery workflow |
Verify
Tools that confirm recovery was actually successful.
| Tool | What It Verifies |
|---|
| Lazarus | Recovery readiness — can your system actually restore? |
Signal Bus
| Tool | Role |
|---|
| Transmission | Moves reliability signals between all components with delivery guarantees |
Common Deployment Patterns
Starting Out: Free Baseline
For teams just getting started with agent reliability.
RadCheck → OCTriage → Lazarus
- Run RadCheck to understand your current state
- Use OCTriage when incidents happen
- Run Lazarus to verify you can recover
Cost: Free.
Production Protection: Core Stack
For teams running production agents that need runtime protection.
Sentinel + Watchdog → Agent911 + FindMyAgent
↓ ↓
Detection layer Response layer
- Sentinel catches runtime anomalies continuously
- Watchdog verifies liveness and escalates
- Agent911 provides the response surface with guided playbooks
- FindMyAgent gives fleet-wide visibility
Best for: Teams with 2+ production agents, recurring incidents, or 24/7 uptime requirements.
Multi-Model Governance: Add SphinxGate
For teams using multiple models or with compliance requirements.
Core Stack + SphinxGate + Transmission
- SphinxGate enforces routing policy and maintains audit trail
- Transmission ensures all signals reach Agent911 reliably
Best for: Teams using fallback chains, multiple providers, or regulated AI usage.
Full Stack: Operator Bundle
All tools. Full coverage across detection, triage, routing, recovery, and verification.
See the Operator Bundle for details and pricing.
Design Principles
Observe, don’t interfere
ACME tools watch your systems. They do not autonomously modify agent behavior, restart processes, or take recovery actions without operator direction. Every action that changes system state is operator-initiated.
This is intentional. Autonomy without understanding creates more incidents, not fewer.
Evidence-first
Every detection includes evidence. Every incident includes a proof bundle. Every recovery includes a readiness check. The goal is that operators always know why a tool is telling them something, not just what.
Deterministic playbooks
Recovery shouldn’t depend on which team member is on-call. Agent911 playbooks give the same guidance to everyone, regardless of familiarity with the system. Same incident type → same playbook → same recovery path.
Trust-first free tier
RadCheck, OCTriage, and Lazarus are free. These tools build the foundation of trust — you understand your system before you buy anything. Paid tools solve the problem at scale.
What ACME Is Not
Understanding what we don’t do is as important as understanding what we do.
- An orchestration framework (we watch your agents, not direct them)
- An autonomous healing system (operators always act)
- A replacement for logging/APM (we’re the reliability layer on top)
- Agent-specific (we work with OpenClaw, LangChain, AutoGPT, custom systems)
Getting Started
