Home · Control Tower · Operations
The platform
The enterprise AI agent platform.
One governed runtime for the whole agent lifecycle — build, automate, govern, observe, and distribute AI agents on a single operating system. Every run flows through the same eight gates and lands in a tamper-evident ledger.
Console → runtime gates → ~22 services → ontology + ledger
One governed runtime
Every run passes the same eight gates.
There is no side door. Whether a run starts from a chat, a workflow, an API call, or a schedule, it crosses identity, budget, guardrails, registry, control-tower, execution, output-guardrails, and audit — fail-closed, in that order.
Identity
Agent IAM — owner, risk tier, expiry, allowed models
Budget
30-day USD caps — hard cap blocks the run (402)
Guardrails
Input guardrails — tone, do-not-say, IO checks
Registry
Allowed models & actions — off-list → 403
Control Tower
Paused or killed → 409 AGENT_PAUSED
Execute
Action Fabric — dry-run → propose → approve → execute
Output guardrails
Output checks — redaction, escalation, format
Audit
Trust Ledger — hash-chained, signed receipts
The platform, end to end
Every capability, in one operating system.
Nine groups, 40+ governed capabilities — the same nine areas your team works in inside the console. Each links straight to its deep-dive.
Agents
Automate
Govern
Observe
Data & Ontology
Cost & Value
Distribute
Admin
Built for everyone, governed by admins
Simple mode for the team. Advanced mode for the platform owners.
One console, two depths. Business users build and chat-test agents in plain English; platform admins drop into the full governance surface — policies, gates, identities, ledgers — without ever leaving the same runtime.
- Simple mode: build → publish → chat-test, no YAML required
- Advanced mode: policy editor, gate config, autonomy modes, registry
- The same guardrails apply in both — there is no “quick” ungoverned path
Agent Studio
skills
Search KBCreate ticketSend email add
Publishedv4 · reliability 96
Where is order #4471?
Shipped, arriving Thursday.tools: Search KB · $0.004
studio ▸ build → publish → chat-test
Four jobs, one runtime
From the prompt to production, and out to every tenant.
Action Fabric
Dry-run›
Refund $240
risk: low · evidence
Proposed›
Close case #4471
risk: med · evidence
Pending approval›
Payout $5,000
risk: high · evidence
Executed
Email customer
risk: low · evidence
actions ▸ dry-run → propose → approve → execute → compensate
Build
Author governed agents in plain English or code.
Compose skills, knowledge, and guardrails in the no-code Agent Studio, or import a signed Solution Pack. Eleven Role Packs — Service, Sales, IT Desk, HR, Finance Ops, Legal Intake, Procurement, Risk & Compliance, Field Ops and more — give every team a governed starting point.
- Skills, knowledge & RAG with audience rules
- Behavior & guardrails: tone, length, do-not-say, escalation
- Agent IAM credential issued on publish
Govern
Policy-as-Code, enforced at runtime — not in a slide deck.
Author testable allow / deny / require_approval rules and simulate them before they ship. Actions become first-class objects with risk tiers and a dry-run → propose → approve → execute → compensate lifecycle. The MCP gateway filters every tool call; oversight modes set a risk floor autonomy can only tighten.
- Simulate amountUsd ≥ 5000 → require_approval before deploy
- Off-allow-list model → 403 MODEL_NOT_ALLOWED
- Action not permitted on object → 409 ACTION_NOT_PERMITTED_FOR_OBJECT
MCP Gateway
mcp ▸ managed registry · kill switch armed
Observability
latency (ms)
Faithfulness0.98
Citation cov.0.95
Safety1.00
promptcontextmodeltoolsapprovaloutput
ai-insights ▸ quality { overall: 100 }
Observe
Score, evaluate, and replay every run.
Per-run quality scoring, evaluation suites as release gates, an A/B simulation lab, a reliability score that gates publish, and an agent debugger with the full run trace. When something looks off, you replay it from prompt to outcome.
- Live run quality { overall: 100 } on the gold suite
- Eval gates block a publish that regresses
- Full run trace: prompt → context → model → tools → approval → output
Distribute
Ship a governed solution once. Deploy it anywhere.
Package ontology, policies, actions and agents into a versioned, signed Solution Pack. Install into any tenant — including air-gapped environments — verified by hash and signature, with the same gates and ledger underneath wherever it lands.
- Signed, hash-verified bundles install across tenants
- Air-gapped & VPC deployment with residency pinning
- Verify offline before install (hashOk: false rejects a tampered pack)
Pack Registry
Fraud TriageExecuted
Payout ApproverHold
Return ReviewerExecuted
KYC ScreenerBlocked
trust-ledger ▸ 1,284,902 actions recorded today
Open by design
Plug into the stack you already run.
Multi-provider model routing, MCP-native tool integration, and your systems of record — Cortex governs the connections rather than replacing them.
Anthropic · Gemini · OpenAI · Mistral · GroqMCP tools & gatewaysYour ontology & data
Security & compliance
Built for the enterprise security review.
Fail-closed gates, RBAC, DLP, agent detection & response, and multi-tenant isolation — mapped to the frameworks your auditors already use.
SOC 2ISO 27001ISO 42001HIPAAGDPREU AI ActNIST AI RMFFINRA
See the whole platform in one runtime.
Walk through the gates, the capability map, and a governed run end to end — then put your own agents under air-traffic control.