Free plan

← All guides

Agent Policies for Multiple Websites: Running a Fleet

Published November 14, 2026 · Reviewed November 2026 · By the Ingenious AI team

The short answer

Every origin publishes its own agent-policy.json — the format has no multi-site document — so running several websites means running a small fleet of files. The fleet pattern keeps that sane: one shared template carrying the family's common posture, per-site decisions where the sites genuinely differ, one register listing every file's version and review date, and a single review sweep that keeps the whole fleet inside one freshness window.

Generate your policy — free · Validate an existing one

What is a policy fleet?

It is a set of policy files under one ownership, sharing a template and a maintenance rhythm while differing where their sites differ. A retail group with four storefronts, a publisher with six titles, a SaaS company with product, docs, and marketing origins — each is a fleet whether it manages itself as one or not. The unmanaged version drifts: one site's file current, another's two redesigns old, postures contradicting each other across sister brands. The managed version costs a register and a calendar entry. The template discipline mirrors how the samples work — decisions written once, applied per site — scaled from one file to a family.

The template: what is shared, what is local

Shared: the structural skeleton (fields, in the family's house order), the default posture where the business model is genuinely common, the standard fences (the committing verbs refused family-wide), the boilerplate data-use clauses that legal has blessed once, and the notice's honest-scope sentence. Local: the site and owner fields by definition; the named rules, because partners differ by site; pace caps, because infrastructures differ; and any posture exception a specific site argues for itself — a flagship store's wider welcome, a sensitive brand's block. The template is a starting document, not a straitjacket: a fleet whose every file is identical has usually stopped making per-site decisions, which is drift wearing uniformity's clothes.

The register: one page that knows the fleet

Four columns per site — origin, policy owner, policy_version, last_updated — plus a fifth for the next review date, is the entire instrument. Its power is comparative: the sweep query "show me the fleet by review date" instantly exposes the file that missed the last two cycles, the way the hotel group in the versioning guide caught its rebrand stragglers. Keep the register where the owners already work — the same tracker as the sites themselves — and let AgentGate's Team-tier change history carry the per-file detail where policies are hosted or generated centrally. A register that requires archaeology to update will not be updated; four columns and a habit beat a platform.

The sweep: reviewing a fleet in one pass

Run the review as one event across the family: fetch every live file, batch-validate the set, and work the exceptions — invalid files first, then stale dates, then posture questions that affect several sites at once (a new ecosystem agent deserving named rules family-wide; a partner renegotiation spanning brands). Per-site owners join for their files' decisions; the fleet owner runs the pass and owns the register. Quarterly suits most fleets, with event-edits flowing through the same channel as single-site work: the site owner edits, the register updates, the sweep catches whatever slipped. The fleet's failure mode is never a dramatic one — it is the quiet multiplication of small staleness, which is precisely what a rhythm exists to prevent.

Worked example: a publisher fleet of six titles finds its rhythm

Fictional example. Callisto Media, a fictional publisher with 6 magazine titles on 6 domains, inherited 6 policies written by 6 web editors across 2 years: postures ranging from open-with-attribution to an accidental block (a sample file committed during a redesign and never revisited). The fleet pass took one week: a shared template from the group's licensing position, per-title rules preserved where editors could argue them (4 titles kept distinct partner rules), the accidental block corrected to the family posture, and a register created in the group's existing tracker. Standing rhythm since: quarterly sweep, 90 minutes for all 6 files including batch validation; event-edits per title as relationships change. The accidental block remains the group's favourite cautionary tale — a policy nobody owned, contradicting the business for 11 months, found by the first fleet review ever run.

The fleet checklist

  1. Template the common. Skeleton, shared posture, standard fences, blessed clauses — decided once, inherited everywhere.
  2. Keep the local local. Named rules, paces, and argued exceptions belong to each site; uniformity beyond the template is drift in disguise.
  3. Maintain one register. Origin, owner, version, review date, next review — four-plus-one columns, kept where owners already work.
  4. Sweep quarterly as one event. Fetch all, batch-validate, fix invalids, then staleness, then shared posture questions.
  5. Route event-edits through the register. Every per-site change updates the family's memory the same day.

Keep reading

Agent Policy for Agencies: Managing Client Policies · Batch Validating Agent Policies: 100 Files per Call · How to Tell AI Agents What They May Do · Spec samples · Conformance · Pricing · All guides

Frequently asked questions

Can one policy file cover several domains?

No. The format is per-origin: each site publishes its own file at its own well-known path, and the file's site field names that origin. A fleet is therefore several files by construction — which is why the fleet pattern's instruments are a shared template and a register, rather than a master document.

Should sister sites share the same posture?

Where the business model is shared, yes — a common default and common fences keep the family's public position coherent and the template meaningful. Where a site genuinely differs (a sensitive brand, a flagship's wider welcome), the file should differ on purpose, with the exception argued and registered. Coherence is sameness where reasons are the same, not sameness everywhere.

Who owns a fleet?

A fleet owner keeps the template, register, and sweep; site owners keep their files' decisions. The split mirrors the work: consistency is a fleet property, relationships are site properties. Small fleets often vest both in one person wearing two hats — acceptable, provided both hats' jobs appear in the calendar.

How do we validate a whole fleet efficiently?

Batch validation: AgentGate's Pro batch API checks up to 100 policies per call, returning per-file errors and warnings in the same distinction the single validator uses. Fetch every live file at sweep time, run the set, and work the exception list. The batch guide covers the workflow, including keeping the fetched copies as the sweep's evidence.

What is the first sign a fleet is drifting?

Spread in last_updated. When one file's review date sits a full cycle behind its siblings, something has already gone quiet — an owner changed, a calendar entry died, a redesign shipped around the file. The register exists to make that spread visible at a glance, before it becomes a posture contradiction an operator finds first.

Generate a policy — free · Validate a policy · What an agent access policy is

← Back