NeMo makes each app safe. Conduct makes the fleet governable.
Five teams pick NeMo Guardrails for their five agents. Each writes its own config.yml. Now IT wants one rule across all five, one audit chain, one kill switch, one spend cap. NeMo doesn't try to solve that — it's a different layer. Conduct does.
Your app safety layer keeps each conversation on-policy. The governance layer keeps the fleet on-policy.
Two layers, one stack
App safety and org governance
are different problems.
App safety layer — NeMo Guardrails
Keeps one application's conversation on-policy. Runs inside the app process, defined per app, evaluated per turn.
- · Colang DSL for rails
- · LLM-based moderation, jailbreak detection
- · Streaming rail evaluation
- · Multimodal (image, audio) safety
- · 20+ third-party detector integrations
Governance layer — Conduct
Keeps the whole fleet on-policy. Sits above every app, every LLM, every agent, every tool call. One rule catalog, one audit chain, one console.
- · Workspace-wide rules across every agent
- · Hash-chained tamper-evident audit
- · HITL approval workflow with Slack + resume tokens
- · Agent identity (
cond_agt_*) scoping blast radius - · Spend caps + budget alerts
- · 8 compliance packs (SOC 2, HIPAA, PCI, EU AI Act, NIST AI RMF, ISO 42001, IRS 1075, OWASP)
The fleet problem
Five apps. Five configs.
One org.
Every team picks the right rails for their app. Good. Now the CISO asks:
Q: How do I block all PII across every agent without editing 5 configs?
A: One workspace rule. Applies to every NeMo app that calls Conduct.
Q: Where do I query every rail decision across all 5 apps for the SOC 2 auditor?
A: One audit surface. Filter by source, decision, rule, date, developer.
Q: Can I pause all agent activity across the fleet if something goes wrong?
A: One kill switch. Every downstream agent respects the workspace policy.
Q: How do I cap total spend across the fleet, not per app?
A: One workspace budget. Enforced at the proxy layer regardless of framework.
Q: When an audit finding lands, can I prove which agents ran what?
A: Hash-chained ledger with agent identity. Every entry attributes to a cond_agt_* token.
Q: How do I ship SOC 2 / EU AI Act evidence from this fleet?
A: Compliance packs export controls-mapped evidence directly from the audit chain.
Wire it up
One Colang input rail.
Every NeMo app becomes fleet-governable.
The conduct-nemo-guard plugin exposes a Colang action that hits Conduct on every user turn. Blocked verdicts short-circuit before the LLM runs. Every decision lands in the workspace audit chain, source-tagged nemo.
Full runnable example: packages/conduct-nemo-guard in the repo.
Observability
See every NeMo rail decision in one view.
Guard Activity filters by source. Pick nemo in the tool filter (or use the deep link below) to see every rail decision from every NeMo app in your workspace.
Already running NeMo in production?
We're hiring two design partners this quarter — teams that already ship NeMo apps and need the governance layer above them. 30-min scoping call.