Ai Kit Ui Livewire Laravel Package

Product Decisions This Supports

• Feature scoping — "chat with our documents" as a shippable feature, not a research project. ai-kit covers the full RAG path (ingestion of PDF/Word/Excel/URL → chunking/embedding → vector retrieval → cited agent responses). For a PM, that's the difference between an epic estimated in quarters and a feature that can be scoped to a sprint or two with a real demo at the end.
• Build vs. buy — the core decision this package informs. The alternative is assembling ingestion libraries, a vector store integration, a retrieval/citation layer, and a UI on top. Adopting a package shifts spend from R&D to integration plus an evaluation/QA budget, and shifts risk from "can we build it?" to "is this dependency good enough and will it be maintained?"
• Surface-by-surface rollout planning. The package ships multiple delivery surfaces, each mapping to a different stakeholder and release path: Livewire UI for end-user product features, Filament panel for internal ops/admin tooling, CLI for batch ingestion and backfills, HTTP API for integrating into existing apps or partner systems, and MCP server for agent/tooling integrations. A PM can sequence value: internal admin tool first (low risk, high learning), customer-facing chat second.
• Roadmap acceleration for AI-assistant initiatives. Teams with a "AI assistant" line item can pull it forward, and teams that had deprioritized it due to ML-hiring constraints can re-open it — the work is PHP/Laravel, not Python/data-science.
• Use cases worth evaluating now: internal knowledge base / HR & policy Q&A, support-deflection over help-center and ticket docs, document-heavy verticals (legal, compliance, procurement, insurance) where cited answers and auditability are the actual product requirement, and "ask our docs" features bolted onto existing Laravel SaaS products.
• Trust and enterprise-readiness decisions. Cited answers are a sales-enablement feature as much as a UX one: citations are the thing that gets an AI feature past security review, legal, and enterprise procurement. Their presence in v0.2.0 is a meaningful signal for regulated or B2B go-to-market.
• Vendor/dependency posture decision. Published under the solution-forest Composer vendor at v0.2.0 — this is an early-stage, single-vendor dependency. That's a decision about acceptable risk and exit cost, not a reason to say no.

When to Consider This Package

Consider it when:

• You are already a Laravel/PHP shop and want to avoid standing up and operating a separate Python/Node AI service — one runtime, one deploy, one on-call rotation.
• The feature you need is retrieval over your own documents with citations, not general-purpose chat, fine-tuning, or model hosting.
• You want a working internal tool fast — the Filament panel and CLI are strong fits for an ops/back-office knowledge assistant that never touches the customer-facing surface.
• Your team has no dedicated ML engineers and no appetite to hire for a single feature.
• You need multiple integration shapes (UI, API, CLI, agent tooling) rather than a single embedded widget.
• You can absorb pre-1.0 churn: pin the version, read the changelog on every bump, keep an integration seam so the retrieval layer is swappable.

Look elsewhere (or wait) when:

• You are at genuine scale — millions of documents, strict latency SLOs, or heavy multi-tenant isolation with per-tenant index guarantees. Validate this explicitly; it is not established by a v0.2.0 release note.
• You need advanced evaluation, guardrails, prompt/version management, and observability as first-class product features — plan to build or add those regardless.
• Compliance, data residency, or procurement requires a mature vendor with an SLA, security documentation, and a support contract. A v0.x community package will not clear that bar today.
• Your stack is not PHP/Laravel, or your team's AI expertise is elsewhere.
• Cost modeling is unresolved — ingestion/embedding volume and per-query token spend can dominate the business case. Do not commit to pricing or usage tiers before you have measured this on your own corpus.
• The roadmap depends on capabilities that are still unproven here: connectors beyond the listed file/URL types, incremental/sync ingestion, permissions-aware retrieval (ACL filtering by document), and answer-quality tuning.

Adoption gate I would set: a one-sprint spike on a representative 200–500 document corpus, measured against a written eval set of 30–50 question/answer pairs. Success = citation accuracy, acceptable "I don't know" behavior, and a per-query cost you can put in a unit-economics model.

How to Pitch It (Stakeholders)

For executives: We can add "ask our documents and get cited answers" to our existing Laravel product using a single Composer package instead of building a retrieval pipeline and an AI service from scratch. That converts a multi-quarter, specialist-staffed initiative into an integration project on the stack we already run and already know how to operate — no new language, no new infrastructure team, no new on-call surface. It comes with admin tooling, an API, and a CLI out of the box, so we can ship an internal version quickly to validate demand before committing to a customer-facing launch. What we are asking for: one sprint of engineering time and a small token/embedding budget to run a measured pilot. What we are not doing: betting the roadmap on it. It is at v0.2.0 from a single vendor, so we pin the version, keep the retrieval layer behind an internal interface, and treat the pilot's evaluation results as the go/no-go.

For engineering: It is Laravel-native and Composer-installable, with the parts we would otherwise build — ingestion for PDF/Word/Excel/URL, vector retrieval, a cited-answer agent loop — plus four integration surfaces (Livewire, Filament, HTTP API, CLI) and an MCP server for agent tooling. Integration cost is low; the real work is on our side: queue/worker capacity for ingestion, secret management for model and vector-store credentials, an evaluation harness, and an abstraction boundary around retrieval so a future swap is a contained change. Please treat v0.x as a moving target — pin it, review each release for breaking changes, and do not let package internals leak into product code. Flag anything in the security review around document access control early; per-user document permissions are the most likely gap and the most likely blocker for a customer-facing release.

For the pilot team: Success criteria are citation correctness on our own corpus, honest "not found" behavior over confident hallucination, and a per-query cost we can price. Instrument from day one.