Cloudflare operations for teams with many sites

A steward for your Cloudflare fleet.

Forty zones, six clients, one support queue. Ask in plain language, approve the fix, read the audit. Nothing writes without you.

Apply for beta How it works Runs on your own Cloudflare token and model key. We don’t sell tokens and don’t mark them up. Pre-launch private beta: free, unfinished, and built on your honest feedback. The trade →
The globeSix ways to read the same traffic
Built for

Every role that touches a zone has its own page. Pick yours.

Proposed, approved, verified, revertable client-a.com
Approval card: Create WAF rule (managed challenge) on WordPress login and XML-RPC endpoints — would have matched about 115 requests of the last 24 hours, broken down by path, network, country and user agent, with a map, and Approve & run or Reject
An approval card, as it appears
The gate

Structurally unable to touch your zones.

Not a policy, not a system prompt. Every function that can change Cloudflare is reachable from the execution core alone, so a prompt injected through a request path or a user agent cannot cause a write.

  1. Impact classified before you see it

    Low, medium or high, decided on the server. A lockdown or an apex DNS deletion needs an acknowledgement, not a click.

  2. Blast radius against real traffic

    The rule is run against your last 24 hours before it exists — paths, networks, countries, user agents — so you see what it would have matched.

  3. Read back from Cloudflare

    “Applied” means the server asked Cloudflare what the setting is now and got the answer it expected.

  4. Reversible, with the payload stored

    Every change lands in the audit trail with the instructions to undo it. One click.

The tour

Six things an agency does with it, on real screens.

Every screen below is the product as it runs. Click any plate to see it full size.

Ask the question the client asked you.

  • “Is it down?”, “Why is it slow?”, “Who is attacking us?” — answered from Cloudflare’s own analytics, as prose with a chart.
  • The query sits beside the answer: Viz, Table, GraphQL, JSON. Nothing is hidden behind the summary.
  • Save it, and ask it again Monday. Every answer shows what it cost, under the answer.

Ten questions, and what happens when you ask them →

Zonesteward chat: a 24-hour health answer on the left, a firewall activity chart on the right with Viz, Table, GraphQL and JSON tabs
An answer and its chart, with Viz, Table, GraphQL and JSON tabs

Ranked, categorised, one click from a challenge.

  • Every source scored and sorted into nine categories, with its country and network beside the address.
  • Block or challenge is a proposal — it goes through the approval card like everything else, with the blast radius counted.
  • Auto-mitigation exists, and is off until you turn it on.
The security tab: attacker counts, category chips, and ranked sources with country flags, networks, probed paths and Block or Challenge actions
The security tab, last 24 hours

Describe what would worry you. It writes the rule.

  • Query, threshold, window and minimum volume, written from one sentence and explained back to you in English.
  • Dry-run against live traffic before it’s armed — a rule that errors or yields no number is refused.
  • Email, Slack, Microsoft Teams, PagerDuty, SMS and signed webhooks.
Alert rules page: a prompt box to build a rule with AI, and two rules — site down on 5xx rate, origin unreachable — each explained in plain words with its threshold, window and channels
Alert rules, written from a sentence

A dashboard for the wall. A chart for any question.

  • Status mix, countries, networks, bandwidth, open incidents and the fleet globe — live, on one screen.
  • Nineteen chart kinds — twelve flat (bars, lines, areas, pies, scatter, histograms, tables, big numbers and more) and seven in 3D, including a globe and a country map. The model proposes up to four for each answer, each with its reason; switch with a click, or ask for fresh ones.
  • Saved charts become dashboard widgets that re-query every minute. Arrange them once and leave it on a screen.

How charts work →

The dashboard: a live globe with request arcs, status code bars, top countries, open incidents, bandwidth and top networks
The dashboard, streaming

One card for a whole client group. One link per client.

  • Group zones by client. A purge, a WAF rule or a rate limit goes to the whole group on one approval card, applied and audited zone by zone.
  • A share link per client — read-only, all their sites or just one, expiring within 30 days — revocable the moment you need it to be.
  • A seven-day digest and incident reports, printable, with CSV for every table.
Create a share link: scope, label, read-only access, and an expiry of 7 days
A read-only share link, expiring within 30 days

Your token. Your model key. Nothing marked up.

  • A wizard scopes the Cloudflare token to exactly what you allow. Anthropic, OpenAI or an OpenAI-compatible endpoint for the model. Both stored encrypted.
  • Every model call itemised — feature, model, tokens in and out, cost — and the cost shown under each answer.
  • Billed to you by the provider, at cost. We don’t sell tokens and we don’t mark them up. A busy month is a few dollars.
AI usage page: cost today, last 7 and 30 days, by feature and by model, and a table of recent calls with their cost
AI usage, itemised
The rest of it

What it can read, what it can change, what it connects to.

What it can read — everything Cloudflare knows about a zone, not just traffic
Edge
  • Edge certificates and expiry
  • DNSSEC, down to the DS record
  • Universal SSL
  • Tiered Cache
  • Cache Reserve
  • Argo Smart Routing
Monitoring
  • Health checks and why they failed
  • Load balancers with pool and origin health
  • Logpush jobs, last push, errors
  • Web Analytics (RUM) setup
  • Cloudflare’s own notification policies
Platform
  • Zone status, nameservers, zone hold
  • Waiting rooms with live queue
  • Observatory results
  • Turnstile widgets
  • Workers routes
  • Access applications
  • Email Routing rules

The full list, by job →

What it can change — always through an approval card, never on its own
  • WAF custom rules, cache rules and rate limits
  • DNS records
  • Zone settings and features
  • IP access rules and lockdowns
  • Managed rulesets
  • Health checks and waiting rooms
  • Cache purge, by URL
  • Pausing a zone
  • Any of the above across a zone group, on one card

Nine steps between a sentence and a change →

What it connects to — your accounts, your keys, your channels
Reads from
  • Cloudflare — any number of tokens, any number of accounts
  • Google Analytics 4
Thinks with
  • Anthropic, OpenAI, or an OpenAI-compatible endpoint you host — your own key, billed to you at cost
Alerts through
  • Email: Resend, SendGrid, Mailgun or SMTP
  • Slack
  • Microsoft Teams
  • PagerDuty
  • SMS via Twilio
  • Signed JSON webhooks
Signs in with
  • Google
  • Microsoft
  • GitHub
  • An emailed link

Every credential is stored encrypted and used only for your workspace. Security →

Rates

Priced per workspace, by zones.

Every tier includes everything. No security feature is held back for a higher plan.

PlanFrom general availabilityIncluded
Studioup to 25 zones $49/mo 3 members. Chat, globe, investigation, approvals, audit, alerts.
Agencyup to 100 zones $149/mo 10 members. Everything in Studio, plus client share links.
Fleet100+ zones Talk to us Unlimited members. Everything, plus priority support.

Free during the beta, then 50% off your first year, locked. Runs on your own Cloudflare token and model key — Anthropic, OpenAI or a compatible endpoint — billed to you by the provider at cost. We do not mark it up.

Private beta · pre-launch

The beta: free for you, honest feedback for us.

Zonesteward is built and in daily use, but unfinished — no public sign-up, no billing, parts still being reworked. Twenty agencies with real fleets get it free and help finish it. Here's the trade.

You get
  • Free for the whole beta
  • Half price for your first year, locked
  • A direct line to the person building it
We ask for
  • Three questions a month, in the app
  • Two twenty-minute calls: month one, month three
  • Your honest feedback

What the beta involves, in full →

Apply for beta →
  1. 01You
  2. 02Your fleet
  3. 03Fit
Who should we write back to?

Takes about two minutes.