Product Decisions This Supports

Cost Optimization for AI Features: Enables leveraging existing MariaDB infrastructure to avoid cloud-based or proprietary vector database costs, ideal for startups or cost-sensitive projects. v0.15.0 raises the floor to MariaDB 11.8+, so the "reuse what we already run" argument now depends on an infrastructure upgrade — worth confirming with infra before this becomes a load-bearing assumption in a business case.
Hybrid Search Capabilities: Supports combining vector similarity search with traditional SQL filtering (e.g., metadata-based queries), enabling use cases like semantic search with category constraints.
Symfony/Laravel Ecosystem Alignment: Facilitates AI integration for teams already using Symfony or Laravel, reducing friction in adopting AI-driven features without requiring new tech stacks.
Build vs. Buy Decision: Provides an open-source, self-hosted alternative to managed vector databases (e.g., Pinecone, Weaviate), avoiding vendor lock-in while maintaining flexibility. Since v0.14.0, VectorDocumentInterface decouples documents from the store implementation, making a later swap to a dedicated vector database a lower-cost decision to revisit — and v0.15.0's index/distance correctness fix means a future migration benchmark compares against a store that behaves correctly, not one with a mis-configured index.
Capacity Planning & Usage Metering: v0.14.0 added count capabilities for stores, giving PMs a first-class primitive for:
- Usage-based pricing tiers and per-tenant quotas (e.g., "1,000 documents on Free, 100,000 on Pro").
- Admin dashboards and "X of Y documents indexed" UI states.
- Ingestion pipeline observability and cost forecasting before committing to infrastructure.
Search Quality as a Product Requirement: v0.15.0 fixes index creation to use the store's configured distance metric (previously an index could be built with a distance that didn't match query-time distance). For PMs this is a quality-bug fix, not a feature — if you shipped a pilot on v0.14.0, re-run relevance/recall evaluation rather than assuming parity.
Roadmap for AI-Driven Products: Enables scalable AI features such as:
- Personalized recommendations (e.g., e-commerce, content platforms).
- Semantic search for documentation, customer support, or internal tools.
- RAG workflows grounded in existing SQL data.
- Anomaly detection or fraud prevention via vector similarity analysis.

When to Consider This Package

Adopt When:

Your vector database needs are moderate (e.g., <10M embeddings) and MariaDB 11.8+ is already deployed or on your near-term upgrade path (11.7 was sufficient through v0.14.0; v0.15.0 requires 11.8 for distance handling — verify before committing).
You prioritize cost efficiency and want to avoid licensing or cloud costs for vector storage.
Your team uses Symfony or Laravel and seeks seamless integration with AI components.
You require hybrid search (vector + SQL filtering) for use cases like metadata-aware semantic search.
You need document counts for quotas, billing tiers, or admin UI — supported natively since v0.14.0.
Compliance or data residency requirements favor open-source, self-hosted databases over cloud-based alternatives.
You are starting fresh in v0.15.0 rather than upgrading a v0.14.0 pilot — the index/distance fix and per-query parsing change mean early adopters should treat this as a re-validation point, not a drop-in bump.

Look Elsewhere When:

You need high-throughput or low-latency performance (e.g., >10K QPS or <50ms latency) for large-scale vector workloads (consider Milvus, Weaviate, or Pinecone).
Your use case demands advanced vector indexing (e.g., HNSW, IVF) or dynamic schema evolution (MariaDB's vector support is nascent).
You require managed services (e.g., auto-scaling, backups) without DIY infrastructure (e.g., AWS OpenSearch, Chroma).
Your team lacks MariaDB expertise or prefers a dedicated vector database with richer tooling (e.g., Qdrant, pgvector).
You are pinned to MariaDB 11.7 or earlier and cannot upgrade — v0.15.0 will not be available to you; stay on v0.14.0 knowingly and plan the upgrade.
You need multi-tenancy isolation or fine-grained access control beyond SQL permissions — note that count is a whole-store primitive, not a per-tenant analytics layer; you will still need to model tenant scoping yourself.

How to Pitch It (Stakeholders)

For Executives:

"This package lets us use MariaDB—already in our infrastructure—as a vector database for AI applications, cutting costs while enabling semantic search, recommendations, and RAG workflows. It's a lightweight, open-source solution that integrates with Symfony/Laravel, avoiding vendor lock-in and licensing fees. v0.14.0 added document counting, which unblocks usage-based pricing tiers and quota enforcement without extra infrastructure, plus a document abstraction layer that keeps our options open if we later outgrow MariaDB. v0.15.0 is a hardening release: it fixes a search-index configuration bug that could have made results less accurate, and makes queries cheaper by parsing the search vector once instead of repeatedly. The one thing to check is that it requires MariaDB 11.8, so confirm that upgrade is on our infrastructure plan before we commit."

Key Benefits:

Cost Savings: Eliminates additional database licensing or cloud costs.
Speed to Market: Seamless integration with Symfony/Laravel reduces development overhead.
Monetization Ready: Native count support (v0.14.0) unblocks usage-based plans, per-account limits, and customer-facing usage dashboards.
Quality Assurance: v0.15.0's index/distance fix removes a latent relevance bug — a risk reduction for anything customer-facing.
Future-Proof: MariaDB's vector support is actively evolving, and VectorDocumentInterface (v0.14.0) reduces coupling to the store itself.
Risk Mitigation: Avoids over-investment in niche vector databases for uncertain use cases.

Watch Item for Leadership: v0.15.0 narrows the supported platform to MariaDB 11.8+. If our fleet is on 11.7, adopting the package now carries an infrastructure dependency — treat it as a coupled decision, not a standalone library pick.

For Engineering/Tech Leads:

*"The symfony/ai-maria-db-store package enables vector storage and retrieval in MariaDB (now v11.8+) for Symfony/Laravel AI applications, offering:

Vector similarity search using cosine or Euclidean distance.
Hybrid queries: Combine vector search with SQL filters (e.g., WHERE category = 'X' AND vector_similarity > 0.7).
CRUD operations: Insert, update, delete, and batch operations for vectors.
Store-level counting (v0.14.0): a count capability on stores, useful for pagination totals, quota checks, and ingestion monitoring without SELECT COUNT(*) on application tables.
VectorDocumentInterface (v0.14.0): documents are expressed against an interface rather than a concrete type, improving testability and easing a future migration to another store.

What v0.15.0 Changes (read before upgrading):

Minimum platform is now MariaDB 11.8, not 11.7 (Distance::Distance requires 11.8). Treat this as the primary upgrade gate; check CI images, local Docker setups, and any read replicas before bumping the constraint.
Index creation now uses the store's configured distance (bug fix). If you created indexes under v0.14.0 with a distance that didn't match query-time distance, results may have been subtly wrong — recreate affected indexes and re-run relevance benchmarks rather than assuming continuity.
Query vector is parsed once per query (perf fix) — reduces per-query overhead; re-measure latency rather than trusting previous numbers on either side.
Version constraints bumped to 0.15; still pre-1.0, so the new interface and count API may shift before 1.0.

Why It's Viable:

Performance: Suitable for moderate workloads (benchmark with your expected vector size/throughput). Treat count as an O(n)-ish operation on large tables — cache it for high-traffic UI.
Tooling: Leverages existing MariaDB backups, monitoring, and SQL tooling.
Extensibility: Built on Symfony's store interface, now with an explicit document contract — swap implementations later without rewriting domain code.

Trade-offs:

Requires MariaDB 11.8+ (verify compatibility with your setup; 11.7 is no longer sufficient).
No built-in GPU acceleration or distributed scaling (works for single-node or read-replica setups).
v0.15.0 is a pre-1.0 minor release with a raised platform floor: pin versions, review changelogs on upgrade, and treat index re-creation as a migration step.

Next Steps:

Confirm MariaDB 11.8+ availability across all environments (app, CI, replicas) before bumping.
Recreate vector indexes on v0.15.0 so the index distance matches the store's configured distance, then re-run relevance benchmarks.
Benchmark latency again post-upgrade (per-query vector parsing change) against a baseline like pgvector.
Adopt VectorDocumentInterface in domain code from day one to keep the store swappable.
Pilot with a non-critical AI feature (e.g., internal search tool) and wire count into the admin view first."*

For Product Managers:

*"This package supports building AI features like semantic search, recommendations, and RAG without requiring a dedicated vector database. It's ideal for:

Prototyping AI-driven products (e.g., personalized search, content recommendations).
Cost-sensitive projects where existing MariaDB infrastructure can be repurposed.
Teams using Symfony/Laravel who want to avoid vendor lock-in.
Plans with usage limits: the count capability (v0.14.0) makes 'X of Y documents used' indicators and tier enforcement straightforward.

What v0.15.0 Means for the Roadmap:

Search quality fix, not a feature: the index now matches the store's configured distance. If you have user-facing quality data from a v0.14.0 pilot, it may be understated — re-baseline before citing relevance metrics to stakeholders.
Cheaper queries: the query vector is parsed once instead of repeatedly, which marginally improves latency and cost per search.
Infrastructure dependency: MariaDB 11.8+ is now required. Add the DB upgrade as a dependency on your delivery plan rather than absorbing it silently inside an engineering ticket.

Considerations:

Performance: Test with your expected dataset size (e.g., <10M vectors); avoid surfacing live counts on every page load.
Integration: Requires MariaDB 11.8+ and Symfony/Laravel; may need custom setup and an infra upgrade.
Scalability: Plan for horizontal scaling if the dataset grows beyond single-node limits.
Stability: Pre-1.0 library with a raised platform floor — expect API and requirement changes between minors; budget an upgrade-review task per release.

Recommendation: Start with a pilot to validate performance and integration before committing to large-scale adoption. Scope the first slice to one feature plus a usage counter, which de-risks both the tech and any pricing model built on top of it. If a pilot already exists on v0.14.0, sequence the MariaDB 11.8 upgrade and index re-creation first, then re-measure search quality before publishing any relevance or SLA commitments."*

Ai Maria Db Store Laravel Package