Key takeaways
- ✓ Blueprint is an official, paid Laravel Boost extension. It adds three agent skills:
planning-filament,reviewing-filament-plansandfilament-security-audit. - ✓ It targets the planning step, where agents most often go wrong (component choice, namespaces, layout, state transitions), not the typing of code.
- ✓ Filament’s own published comparison shows that a strong 2026 model already writes a decent v5 plan without Blueprint. Blueprint adds decision gates, exact commands and doc links, concurrency-aware state machines and a verification plan.
- ✓ It’s worth it for teams building non-trivial Filament features with agents, and optional for small CRUD work, where the free Boost skill is enough.
When Filament v5 launched in January 2026, the more interesting announcement was a paid product: Filament Blueprint, “a tool that helps AI coding agents produce better implementation plans for Filament v4 and v5 projects.” Blueprint 2.x now targets Filament 5. This review covers what it actually does, what the evidence shows, and who should pay for it.
A note on method: this review is based on Filament’s documentation, the official before/after example plans, and community testing. It doesn’t claim benchmark numbers we haven’t measured. At the end, we explain how to evaluate Blueprint on your own codebase in an afternoon.
What Blueprint is (and isn’t)
Filament already offers a free Agent Skill for Laravel Boost. It teaches implementing agents (Claude Code, Cursor, Copilot and others) to write idiomatic Filament code, and Boost adds documentation search. The Filament AI docs describe that skill as designed “primarily for implementing agents”.
Blueprint goes after a different problem: the plan. As the docs put it, “the quality of AI-generated code depends heavily on the quality of the plan… A vague plan leads to vague code.” Blueprint ships three skills:
| Skill | Job |
|---|---|
planning-filament | Produces an implementation-ready spec: models, resources, forms, tables, actions, authorisation, tests |
reviewing-filament-plans | Checks an implementation against the plan and reports Conforms / Deviates / Blocked / Not verified for each requirement |
filament-security-audit | Scans a Filament codebase for known misconfigurations (authorisation, uploads, XSS, query scoping, dependency CVEs) and writes a remediation plan |
It’s not a code generator, a UI builder or a replacement for the free Filament skill. It’s designed to be used by a planning agent, so the implementing agent’s context stays free for code.
Installing it
After buying a licence:
composer config repositories.filament composer https://packages.filamentphp.com/composer
composer config --auth http-basic.packages.filamentphp.com "YOUR_EMAIL_ADDRESS" "YOUR_LICENSE_KEY"
composer require filament/blueprint --dev
php artisan boost:install # select Agent Skills → filament/blueprint (skills)Check that your agent’s skills directory contains planning-filament/SKILL.md, reviewing-filament-plans/SKILL.md and filament-security-audit/SKILL.md.
The workflow
1. Plan (in your agent’s planning mode):
Using the planning-filament skill, create a Filament Blueprint for a SaaS invoicing
system using this application's Filament version. Write the plan as Markdown to
blueprints/invoicing.md. ... Flag unresolved business choices.2. Resolve the open decisions yourself. This step matters. More on it below.
3. Implement (switch out of planning mode):
Implement blueprints/invoicing.md in this application checkout. Follow the plan
and run the relevant tests. Report what was implemented and any checks that
could not be completed.4. Review the implementation against the plan with reviewing-filament-plans, without changing code.
5. Optionally audit with filament-security-audit, writing findings to SECURITY_AUDIT.md.
What the official before/after example shows
Filament publishes two plans for the same invoicing prompt, written by the same model (GPT-6 Astra in Amp) without Blueprint and with Blueprint. Reading both is the best way to judge the product. Our takeaways:
The “without” plan is not bad. This is worth saying plainly. It uses correct v5 conventions (Filament\Schemas\Schema, Filament\Actions, Repeater::make('items')->relationship()), maps flows to resources and relation managers, and even warns that “hiding buttons is not authorization.” Frontier models in late 2026 know Filament v5 well, especially with the free Boost skill installed.
The “with” plan is more operational. The Blueprint version:
- Opens with decision gates (G1–G6) that block implementation until someone answers ownership, currency and tax, deletion and correction, delivery, payment and numbering questions. Assumptions are labelled “proposed baseline, pending business confirmation.”
- States exact versions, artisan commands and namespaces, plus documentation links for each component to fetch.
- Specifies models as schema tables, with types, casts and invariants (for example money stored in minor units).
- Models state transitions as a guarded state machine: Draft → Queued → Sending → Sent → Partially paid / Paid, with
lockForUpdate, version checks and idempotent job claims (“Duplicate jobs must result in one atomic claim”). - Ends with a verification plan: risk/outcome pairs mapped to test entry points, including precision and stale-version cases.
So Blueprint doesn’t turn a bad plan into a good one. It turns a good plan into one an implementing agent can follow without making up business rules or skipping concurrency and security details. Those are exactly the places where agent-built admin features cause production incidents.
Community testing points the same way. Laravel Daily’s video “I’ve Tried Filament Blueprint with Claude Code (real Upwork project)”, linked from their This Week in Laravel roundup, is a useful independent look.
Where Blueprint helps most
- Features with workflows and money. Invoicing, approvals, subscriptions, refunds: anything with state transitions and concurrency.
- Cheaper or faster implementing models. A precise plan lets a cheaper model implement reliably, while an expensive model plans once.
- Teams. A saved
blueprints/*.mdfile is a reviewable spec that product owners can sign off. The gates turn “the AI decided” into “we decided”. - Security posture. The audit skill is a structured checklist of Filament-specific mistakes, such as unscoped queries, unsafe uploads and missing policies. You could write this checklist yourself, but most teams don’t.
Where it’s overkill
- Simple CRUD. “Add a Tag resource with name and colour” doesn’t need gates. The free Filament skill plus
make:filament-resource --generateis enough. - Solo developers who already write detailed specs. You’re doing Blueprint’s job by hand.
- Non-agent workflows. If you don’t use AI agents, Blueprint does nothing for you.
Pricing and licensing
Blueprint is sold through Filament Packages. Key terms from the licence:
- Per developer. “Each developer who uses Blueprint must have their own license.” Seats can’t be shared.
- Unlimited projects and unlimited local and staging environments per licensed developer.
- One-time purchase (a year of updates, then keep the version you have) or subscription (updates while active).
- Business use only, and you don’t get the source code.
Check the checkout page for current prices. Budget it per seat, like any developer tool.
How to evaluate it yourself in an afternoon
Don’t trust anyone’s review, including this one. Run a small A/B on your own code:
- Pick a real, medium-complexity feature from your backlog (a workflow with at least two state transitions).
- Run A: plan with your usual agent plus the free Filament skill. Implement. Count correction cycles (times you had to intervene) and defects found in review.
- Run B: plan with
planning-filament, resolve the gates, implement, then runreviewing-filament-plans. Count the same things. - Compare the time to a merged PR and the number of post-merge fixes over the following two weeks.
If Run B saves more than about an hour on one feature, the seat has probably paid for itself.
Verdict
Blueprint is a well-designed, narrowly focused product. It isn’t magic. Modern models already write competent Filament v5 code, especially with Laravel Boost’s free Filament skill. What Blueprint adds is discipline: explicit business decisions, exact APIs, concurrency-aware workflows and a verification plan, plus a review and audit loop.
- Buy it if your team regularly ships non-trivial Filament features with AI agents.
- Skip it for small CRUD panels or if you don’t use agents.
- Either way, install the free Boost skill and write good agent rules (see Claude Code and Cursor on a Filament codebase: rules that work).
FAQ
What is Filament Blueprint?
An official, premium Laravel Boost extension from the Filament team. It gives AI agents skills to write detailed Filament implementation plans, review implementations against them, and audit Filament apps for security misconfigurations.
Does Filament Blueprint work with Filament v5?
Yes. Blueprint 2.x supports Filament v5. It was launched alongside v5 in January 2026 with support for v4 and v5 projects.
Is Filament Blueprint free?
No. It’s licensed per developer, as a one-time purchase with a year of updates or as a subscription. Filament’s basic Agent Skill for Laravel Boost is free.
Which AI agents work with Blueprint?
Any agent supported by Laravel Boost’s Agent Skills, including Claude Code, Cursor and GitHub Copilot. Filament’s own example uses GPT-6 Astra in Amp.
Sources and further reading
