Lens For Laravel Laravel Package

Product Decisions This Supports

• Ship an accessibility remediation capability without building one. Lens for Laravel bundles scanning, AI-drafted fixes, human review, and apply — a build-vs-buy answer that favors buy/adopt for Laravel teams, letting engineering focus on product rather than an LLM remediation pipeline.
• Prioritize "fix-all" workflows on the roadmap. v3.5.0's bulk apply (waits for pending generations, writes every ready suggestion, per-item failure reporting, one rate-limit retry, applied/failed summary) turns a queue of accessibility findings into a batch operation. This is a workflow-level feature PMs can put in front of customers as a productivity story.
• Put AI spend and quality on a dial. The new optional LENS_FOR_LARAVEL_AI_MODEL override lets you pin one explicit model per provider (e.g., gpt-5.6-luna for OpenAI), with empty meaning "unchanged SDK default." That is a product lever for cost control, output quality, and support reproducibility — and a decision PMs should own, not leave buried in ops.
• Improve reliability across providers as a roadmap item, not a one-off bug. The temperature change (no longer pinning to 0; provider default used) unblocks models without temperature support. This matters for provider-portfolio decisions: it widens the set of viable models, including Ollama/local.
• Make governance visible in the product surface. The dashboard scanner card now reports the configured provider with either the default SDK model or the manually configured model name, hidden while AI Fix is unavailable. That is a supportability and trust feature — useful for internal stakeholders and customer-facing transparency.
• De-risk upgrades. v3.5.0 requires no migrations; existing configs need no changes because an empty override preserves prior behavior. This supports a "ship the upgrade in a normal sprint" decision rather than a migration project.
• Use cases: marketing/CMS sites, SaaS dashboards, regulated or public-sector properties where accessibility findings need remediation at volume, and teams wanting a local-model path (Ollama) for data-handling reasons.

When to Consider This Package

Consider adopting when:

• You are on Laravel and want accessibility scanning plus AI-assisted remediation in the same stack.
• You want human-in-the-loop review — the bulk apply explicitly uses each item's reviewed or edited code, which fits compliance-aware workflows.
• Provider flexibility matters: explicit model pinning, Ollama fallback via ai_ollama_model, and the v3.4 additional-providers work reduce single-vendor dependence.
• You want batch remediation throughput (Fix All queues) without losing per-item failure reporting or rate-limit handling.
• You value dashboard-level transparency about which model is doing the fixing, and you need localized UI (EN/PL/ES/FR/DE as of this release).
• You want low-friction upgrades: no migrations, backwards-compatible config.

Look elsewhere when:

• You are not on Laravel/PHP, or need a framework-agnostic or hosted-only accessibility service.
• Your security/procurement process cannot accept sending page content to third-party AI providers, and you cannot run Ollama or an equivalent local model.
• You need deterministic, rule-based fixes with zero LLM exposure — AI-generated code changes may not satisfy strict audit requirements without additional controls.
• You require contractual SLAs, vendor support tiers, or certifications (e.g., SOC 2, DPA coverage) that a community-style package may not provide.
• You need the override to be a supported, guaranteed contract rather than "at your own responsibility" — the model pin is explicitly user-owned risk, so teams wanting a vendor-backed model matrix should treat that as a gap.
• You are air-gapped with no practical path to a hosted or local model endpoint.

How to Pitch It (Stakeholders)

For executives: This release makes AI-assisted accessibility remediation more controllable and more efficient. Teams can now explicitly choose the AI model (cost and quality control), see which model is active in the dashboard, and fix entire queues in one action with clear success/failure reporting. There are no database migrations and existing setups keep working, so the upgrade is a routine dependency bump rather than a project.

For engineering: v3.5.0 adds an optional ai_model config key mapped to LENS_FOR_LARAVEL_AI_MODEL; empty preserves the SDK default, and for Ollama the explicit override takes precedence over ai_ollama_model. Bulk apply shares a single apply request implementation with the single-apply flow, locks queue navigation during a run, and cancels on modal close. The accessibility agent no longer forces temperature to 0, fixing provider errors on models without temperature support. Passthrough tests cover model routing, precedence, bulk-apply wiring, and escaped model names; translations are included for five locales.

For risk/compliance: Human review remains mandatory before apply; bulk apply uses reviewed/edited code per item, reports failures individually, and performs a single rate-limit wait-and-retry. Model selection is explicit and visible, which supports change management and incident triage — but note the override is documented as "at your own responsibility," so model choice should be captured in your operational runbook.