Apex
The AI command center — a natural-language control plane over the entire platform, wired into 103 real capabilities with full permission and approval enforcement.
At a glance
| Route | /apex (legacy alias /command-center) |
| Group | Operate |
| Page permission | command_center |
| Screens | 11 |
What it is
Apex is a natural-language control plane over the entire platform. You describe what you want in plain English; Apex resolves it into one or more platform capabilities, gathers the data, reasons over it, and either answers the question or proposes an executable plan.
It is the single largest subsystem in the product.
Not a chatbot bolted onto a dashboard
Apex is built as a capability registry, not a prompt. 103 capabilities are registered, each a typed, permission-aware binding to a real platform function. 420 key aliases map the many ways a human might phrase a request onto those capabilities.
This is why its answers are trustworthy: it is calling the same functions the dashboard calls, under the same permission checks.
Who it is for
Everyone. It is deliberately the lowest-friction entry point in the product.
- A finance analyst who will never learn the dashboard can ask a question and get a grounded answer.
- A platform engineer can execute multi-step remediation faster than clicking through modules.
- An executive can ask for a summary without opening a twelve-panel report.
How it works
Grounding, not recall
The grounding layer pulls real workspace data so the model reasons over facts rather than recollection. This is what makes the numbers trustworthy — Apex is not remembering what your spend looked like, it is fetching it.
Reads execute. Writes never do.
The write safety contract
Write operations are never executed directly. They are returned as a plan that you must
confirm by typing the literal phrase APPROVE, and that confirmation is authorised by a
single-use token so a replayed request cannot fire the same change twice.
Multi-step pipeline execution is additionally gated by the plan's pipeline execution flag, restricting it to Business and Enterprise.
Memory and continuity
Working memory, conversation memory, and an organisational knowledge base give the agent continuity across turns and sessions — so you are not re-explaining your tagging convention every morning.
Model routing
Apex routes across multiple providers, selecting per task class and falling back when one is unavailable. Every call is metered against your workspace's credit ledger, so AI spend is itself an auditable cost line.
Features
The eleven screens
Apex is not one page. Each screen is a distinct route under /apex.
| Screen | Route | What it does |
|---|---|---|
| Chat | /apex/chat | The conversational surface — streaming, threads, export |
| Overview | /apex/overview | Agent activity, usage, and health at a glance |
| Cost Incidents | /apex/incidents | Persistent RCA for spend spikes |
| Knowledge | /apex/knowledge | Workspace documents the agent can search and cite |
| History | /apex/history | Every run, with per-step events and cancellation |
| Commitments | /apex/commitments | Agent-assisted commitment analysis and negotiation |
| Evaluations | /apex/evaluations | Answer-quality measurement |
| Permissions | /apex/permissions | Which capabilities the current role may invoke |
| Customization | /apex/customize | Behavioural and surface configuration |
| Instructions | /apex/instructions | Response style profiles and standing instructions |
| Readiness | /apex/readiness | Pre-flight check on connections, data, permissions |
How to use it
Open Readiness first
It reports whether cloud accounts are connected, whether spend data has been ingested, and whether you hold the permissions the agent would need. Resolve anything it flags before relying on answers.
Set your response style
In Instructions. Choose executive for leadership summaries or guided for teams new to FinOps. Add any workspace-specific standing instructions here.
Load organisational context
In Knowledge — cost allocation conventions, naming standards, tagging policy. Anything the cloud APIs cannot tell the agent.
Start with a read-only question
In Chat. For example: "what are the top five services by cost this month?" Confirm the answers match the dashboard before trusting anything more consequential.
Move to an action request
For example: "run a waste scan and return high-severity opportunities." Review the returned plan carefully — it states exactly what will be executed.
Type APPROVE to execute
Watch the run in History, which streams per-step events and permits cancellation mid-flight.
Review Permissions
Confirm which write capabilities your role may reach, and tighten if necessary.
Use Incidents whenever spend spikes
Trigger an investigation, review the evidence timeline, assign an owner, request remediation, resolve. The incident and its resolution are remembered for next time.
Automate the routine
In Scheduled Tasks — for example a Monday-morning brief on the previous week's movement.
Track consumption
In Evaluations and Membership → Apex AI Credits. Set guardrails before the allowance is exhausted.
Why it matters
Apex removes the training requirement that makes most FinOps tooling fail on adoption. The people who most need cost answers — finance analysts, engineering managers, executives — are precisely the people least likely to learn a dense dashboard. Apex lets them ask.
For engineers, it collapses multi-module workflows into a sentence while preserving every permission check, approval gate, and audit entry the manual path enforces.
The Cost Incident Center in particular converts spend-spike investigation from an ad-hoc scramble into a repeatable process with institutional memory.
Connects to
- Reads from every Analyze module
- Writes through Govern
- Incidents connect to Opportunity Queue and Approvals
- Credits are managed in Membership
- ChatOps reach is configured in Integrations