Endcap
A drop-in, wire-compatible replacement for Bloomreach Discovery: point your unmodified storefront at your own infrastructure and search keeps working — then own it
The Problem
Search and merchandising is the highest-leverage software on an e-commerce site — and it usually runs on a hosted platform billed by traffic and GMV. Every query, ranking decision, and click flows through someone else's cloud, along with your catalog and your shoppers' behavioral history.
The integration runs deep: feed pipelines, pixel beacons, suggest boxes, recommendation slots, and a merchandising workflow your team already knows. Replacing the platform has always meant a re-integration project, which is exactly why the bill keeps getting paid.
What It Is
Endcap speaks Bloomreach Discovery's wire protocols — the core search API, suggest, recommendation widgets, the tracking pixel, and catalog feed ingestion over REST and gRPC — closely enough that an unmodified storefront integration pointed at endcap instead of the hosted endpoints keeps working. The same account credentials your storefront already sends are configured on the endcap side; the swap is a config change, not a rewrite.
Nothing on the wire was guessed. Every contract was verified against Bloomreach's own published API specs and documentation — vendored and cited in the repository — and a black-box test suite exercises every endpoint. Where older integrations expect legacy response keys, endcap emits the documented shape plus the legacy keys, because extra keys don't break clients and missing ones do.
Key Features
A real search engine
An in-memory inverted index with BM25 ranking, facets, prefix suggest, edit-distance spell correction, synonyms, and CJK-aware tokenization — one Go binary, zero runtime dependencies, no external services.
Merchandising, self-serve
A merchandising console for pin-to-slot, boost, bury, block, banner campaigns, synonyms, and redirects — with live preview showing exactly which products got pinned, boosted, buried, or blocked.
It learns from your traffic
The pixel beacons your storefront already sends power analytics, event-trained ranking, mined did-you-mean corrections, co-view recommendations, and semantic rescue for zero-result queries.
Experiments built in
Deterministic A/B bucketing with treatment-vs-control lift reporting — plus auto-optimize, which shifts traffic toward the winning arm while the experiment runs.
Operate it yourself
Multi-tenant accounts, leader/follower replication for high availability, hourly snapshots, Prometheus metrics, and plain-JSON state with no database. Back up a directory; that's the ops story.
Sized for real catalogs
Measured on a million-product catalog: roughly 574k records per second of feed ingest and ~170ms p95 keyword search — on one modest node.
The Swap
Cutover is a documented playbook, not a leap of faith: endpoint mapping, credential and account handling, catalog migration through the same feed API you already run, a staged rollout with rollback, and a post-cutover verification checklist. Because the storefront integration doesn't change, you can point a staging environment at endcap, compare responses side by side, and flip traffic when you're satisfied.
After cutover you own the deployment outright: your catalog, your events, your ranking, your infrastructure — with no per-call metering between your shoppers and their search results.
Talk to sales about Endcap
Tell us what you're running today and what it costs you, and we'll walk through what pointing your storefront at your own search stack looks like.
