Fewer vendors,
one substrate.
Most of the boxes you bolt onto a Laravel app — a cache, a queue, a search cluster, a metrics vendor, an object store — are features of a data layer and a runtime, not separate products. Elyra composes them out of the SQL engine and the server you already run.
This page is deliberately honest: each capability is marked Shipping, Partly shipping or Roadmap.
Search is a chapter, not a product
Full-text (MATCH … AGAINST with a persistent inverted index), faceted search (FACET), and vector search (VECTOR type + similarity) are the same engine that already stores your rows. SQL Anywhere unifies FTS5 + faceted + vector + hybrid in the embedded variant.
Cache, queue & pub/sub
The queue, cache and pub/sub contracts are conformance-tested (CI-verified) in SQL Anywhere, and Askr ships durable L2 backends for all three over SQL Anywhere (--features sql-backend + ASKR_*_DB) — durable, replicated, at-least-once — with L1→L2 write-through, backlog autoscaling on the L2 queue, and a Laravel broadcasting driver. This completes the Redis-free Laravel surface: session + cache + queue + broadcasting, behind the unchanged askr_* / Laravel APIs.
The durable tier is opt-in — now shipped precompiled in the -full tarball / Docker image (no building from source), inert until ASKR_*_DB is set. Making it the default and publishing head-to-head Redis-vs-SQL-Anywhere benchmarks is the remaining polish.
Observability on your own stack
SQL Server ships real OLAP (columnar scans, zone-maps, benchmarked) — the right home for metrics and logs. Askr streams a structured per-request log into ElyraSQL over the MySQL wire (--features observ), plus a periodic metrics-rollup table (p50/p95/p99, inflight), all queryable with plain SQL in Conductor. Askr also exports OpenTelemetry traces (root http.request + php.execute + response.build spans) over OTLP/gRPC to Jaeger/Tempo/the OTel Collector.
Framework telemetry emit and a first-class Conductor observability viewer (dashboards over the SQL rollup) are the remaining pieces.
Deploy is provisioning, pointed remote
Grove already provisions a whole environment — PHP, Node, database, TLS — from one binary. Deploy is that same job aimed at a remote host in production mode.
Remote/production deploy, release history and rollback don't exist yet; Conductor has no deploy cockpit. (And the glue CLI is grove deploy — not rata, which is the desktop Framework CLI.) The design is now agreed and public — the deploy contract fixes the release layout, the step semantics and the rule that a rollback restores code, never data — and it is sliced: releasing to a box that already runs Askr first (elyra-17), provisioning a bare box after (elyra-20). Askr already owns the reload, so v0.1 is the missing middle, not a Forge clone.
Object storage, same substrate
SQL Anywhere already replicates and streams to S3-backed bottomless storage, and Askr is the HTTP server — the reusable foundations are there.
No chunk/blob store and no S3-compatible API exist yet. A chunk-store + metadata could live on SQL Anywhere (inheriting replication) with Askr serving the S3 face — greenfield, but on a reused substrate.
The pattern
A capability lives in the substrate (the SQL engine that's already embeddable and replication-ready) and is served by the runtime (the server that already speaks HTTP). New surface area, reused foundations — not a new binary per box on the diagram.