The Multi-Agent OS
for SRE & DevOps
100+ autonomous AI agents. 13 safety gates. 300+ integrations.
Built for the teams who get paged at 3 AM.
Pre-seed · $2M round · Q3 2026
100+ autonomous AI agents. 13 safety gates. 300+ integrations.
Built for the teams who get paged at 3 AM.
The average SRE stack costs $300K–$1M/year. PagerDuty fires 500+ alerts per week. Mean time to resolve: 47 minutes. Meanwhile, engineers burn out and quit.
The incumbents were built for dashboards and humans on-call. The world has moved to AI agents — but the infrastructure hasn't.
Frontier model costs dropped from $60/M tokens to $0.60. Multi-agent systems that were economically impossible in 2024 are now viable at scale.
Claude, Gemini, and GPT-4 can now follow multi-step playbooks, parse logs, and execute remediation with human-level judgment.
Datadog and PagerDuty bolt on AI. We built for it from day one. The category is "Agentic SRE" and it doesn't exist yet.
100+ AI agent personas organized into 16 specialized teams. 78 active in the production orchestrator. Each agent has a defined role, expertise domain, and tier-appropriate model assignment.
Incident Commander, Auto-Remediation, Escalation Manager, and SLA Guardian work together to resolve incidents without waking humans.
Pattern Miner, Correlation Engine, and Root Cause Analyzer process metrics, logs, and traces to pinpoint issues in seconds.
Policy Enforcer, Quota Manager, and Compliance Officer ensure every agent action meets organizational standards.
React + TypeScript + Socket.IO. Full incident management, agent activity, integration health, cost tracking, and team dashboards. Every metric live via WebSocket.
Manage incidents, query agents, drill into metrics, manage runbooks, control escalations, and configure integrations — all without leaving the terminal.
Production bash agent for Linux, macOS, Windows. Collects CPU, memory, disk, network, GPU, Docker, and auto-discovers running services. Zero dependencies.
| Capability | Datadog / PagerDuty | Single-agent wrappers | Nova AI Ops |
|---|---|---|---|
| Multi-agent orchestration | ✗ | 1 agent | 78 agents, 16 teams |
| Multi-model routing | ✗ | 1 model | 4 providers, 3 tiers, circuit breakers |
| Safety gates | Basic RBAC | ✗ | 13 production gates |
| Post-remediation verification | ✗ | ✗ | T+5m/1h/24h probes, auto kill-switch |
| Digital-twin dry-run | ✗ | ✗ | Simulate before execute |
| Consensus voting | ✗ | ✗ | Multi-agent proposals, human override |
| Native integrations | 700+ | ~20 | 300+ (323 connector files) |
| Pricing (10-seat team) | $200K+/yr | $5K–$20K/yr | $3,480/yr Standard |
We've analyzed 75 competitors. The market splits on two axes: AI-native vs bolt-on, and single-agent vs multi-agent OS.
78 specialized agents across 16 teams, each with defined roles, expertise, and model tiers. Competitors would need to build the orchestrator, the safety layer, and the agent library simultaneously.
Kill switch, prompt-injection defense, cost breaker, SLO gate, tenant isolation, ground-truth verifier, consensus arbiter, simulation engine, counterfactual replay, dangerous command guard, blast radius guard, context redactor, prompt egress scanner. Every gate has API routes, core logic, and audit tables.
323 native connector files covering AWS, Azure, GCP, Kubernetes, Datadog, PagerDuty, Splunk, Terraform, and 300+ more. Each integration makes the next one more valuable.
Organizations → Workspaces → Teams (hierarchical), time-bound permissions, SAML 2.0 + OIDC with PKCE, SCIM provisioning, MFA. Not bolted on — it's in the schema and middleware.
Routes across 4 providers (Anthropic, Google, DeepSeek, OpenAI) with per-agent tier assignment, circuit breakers (5 failures → 60s open), and automatic fallback chains. Every LLM call logged with provider, model, tokens, latency, and cost. Designed for 5–50x lower inference cost vs single-model wrappers.
Every agent action passes through multiple gates before it touches production. Every decision is logged for audit.
Three-scope emergency brake: agent, tenant, or global. Arm/disarm with full audit trail.
~20 regex patterns, severity ladder (none→critical), quarantine table for forensics.
Configurable spend ceilings per tenant. Auto-halt on breach. Budget check before every LLM call.
Policy matrix: risk level × budget remaining. Blocks risky actions when error budget is depleted.
Defense-in-depth: org_id on all tables, middleware verification, violation recording.
T+5m, T+1h, T+24h post-remediation probes. Auto kill-switch on critical regression.
Multi-agent proposals, resolution voting, human override for escalated ties.
Digital-twin dry-run: service graph snapshot, step handlers, risk scoring before execution.
Replay past incidents with alternative actions to validate agent decision quality.
Pattern-match destructive commands (rm -rf, DROP TABLE, kubectl delete) before execution.
Estimate impact scope before any remediation. Block actions affecting too many services.
Strip secrets, PII, and credentials from agent context before LLM calls.
Not a wrapper around one frontier model. An inference orchestration layer that routes the cheapest capable model per task and fails over automatically.
Opus (8192 tokens) for Incident Commander and RCA. Sonnet (4096) for most agents. Haiku (1024) for scribe and summary. Per-agent tier assignment, not one-size-fits-all.
Anthropic → Google Gemini → DeepSeek → OpenAI. Circuit breaker opens after 5 failures, 60s reset. No single provider can take down the system.
Provider, model, input/output tokens, latency, cost estimate — recorded to SQLite and broadcast via WebSocket in real time. Full audit trail for compliance.
File: backend/src/core/aiModelRouter.js · backend/src/core/llmTelemetry.js
Annual discount: 20%. Multi-year: 22% (2yr), 30% (3yr) on Enterprise.
The SRE and DevOps tooling market is massive and fragmented. The average mid-size team spends $300K–$1M/year across observability, incident management, and automation.
Nova replaces or consolidates 3–5 tools in the stack. The wedge is incident response; the platform absorbs observability, runbooks, and cost management.
SEO moat: 75 competitor comparison pages, 2,289 technical blog posts, owning "Agentic SRE" SERP.
SRE with 10+ years at JPMorgan Chase and the US Navy. PhD, MSc, MBA. Built the entire platform.
Brand, content, and go-to-market strategy. Driving the PLG motion and community growth.
M.Eng, PMP. Enterprise partnerships, strategic planning, and investor relations.
AI strategy, business development, and key account management.
Core agent development, model routing, and safety gate implementation.
Hiring: 3 planned — Senior Backend Engineer, ML Engineer, DevRel.
This is not a demo. Every layer has real code, real tests, and real infrastructure.
SQLite (WAL) with Postgres dialect translator. org_id on all tables. 7 migration files. Audit ledger, LLM event log, isolation violation tracking.
Full OTEL SDK with auto-instrumentation. Jaeger/OTLP exporters. Prometheus metrics. Structured logging. Self-monitoring.
Kustomize manifests with dev/staging/prod overlays. HPA auto-scaling. Nginx blue-green traffic switching with health checks. GitHub Actions CI/CD.
Confluent, Databricks, Google for Startups, NVIDIA Inception, MongoDB for Startups, Redis for Startups, AWS Activate.
GDPR and CCPA live. HIPAA BAA on request. Controls modeled on AICPA Trust Services Criteria and ISO 27001 Annex A. DPA available.
TLS 1.3, AES-256 at rest, SAML 2.0 SSO (Okta, Azure AD, Google, JumpCloud), SCIM, MFA, API keys with prefix-based lookup, IP whitelisting.
Pre-revenue by design — PLG requires product-market fit before monetization. 69 organic signups validate demand.
→ First paid conversions targeted Q3 2026Founder built the entire 2,000+ file codebase. Team is lean but the architecture is production-grade.
→ 3 hires planned with round proceeds44 test files for 2,000+ source files. Safety gates and core path are the priority.
→ Prioritizing safety gate + integration tests in Q3Datadog and PagerDuty have the data. But their architecture is dashboard-first, not agent-first. Retooling is a multi-year effort.
→ 13 safety gates + 78 agents = 18+ months head start3 hires: Senior Backend, ML Engineer, DevRel. Accelerate agent library, test coverage, and integrations.
PLG infrastructure, content, community, and first enterprise pilots.
SOC 2 completion, multi-region deploy, and compliance certifications.
The Multi Agent Operating System for SRE, DevOps, and Reliability Teams.
13 safety gates. 300+ integrations. Production-grade.