status: deferred
timestamp: 2026-06-20
tags: [search, orama, vector, hybrid, deferred, future]
Orama Cloud
Deferred — in-browser vector + keyword search, revisit if needed
Orama Cloud
Status
Deferred. Documented as a candidate replacement for Pagefind on the smaller / static sites if vector search ever becomes a valuable addition to the family’s search UX (e.g. semantic-similar-posts, embeddings-driven “more like this” panels). Not adopted in 2026-06-20 Batch 11 — user direction was “i am not sure”, interpreted as don’t add now, document for later.
If / when this flips to status: active, the swap target is
Pagefind on small/static sites; Algolia stays on the
large-corpus sites because Orama’s free tier (10K docs) is below
those sites’ index sizes.
Role (when adopted)
In-browser hybrid search — runs as a WASM module in the visitor’s tab, indexing pulled at first interaction; falls back to the Orama Cloud hosted instance for vector / embeddings queries. Different shape from Algolia (hosted full-text) and Pagefind (build-time keyword) — adds semantic retrieval and the option of a client-side privacy-first index.
Free tier
- 10,000 documents stored
- 5,000 searches / month
- Hybrid retrieval (BM25 keyword + vector embeddings)
- All client SDKs (React / Vue / Svelte / vanilla JS) on the free tier
- Self-hosted Orama (
@orama/orama) is OSS — escape hatch if the hosted free tier ever tightens - No card required at sign-up
Card / subscription required?
NO. Free-tier sign-up is GitHub-OAuth-only. No payment method requested.
When to revisit
Promote deferred → active (replacing Pagefind on small/static
sites) when any one of these holds:
- The family ships an embeddings-driven feature (e.g.
“semantically similar posts” on
oriz-blog) that Pagefind’s keyword index can’t serve. - Pagefind’s bundle size on a large site crosses the Cloudflare Pages static-asset budget.
- Orama Cloud’s free tier expands meaningfully or its hybrid-retrieval becomes a clear UX win the user wants on every site.
Until one of those holds, Pagefind stays — its zero-infra + zero-network-call posture beats Orama on simplicity for small corpora.
Alternatives
- Pagefind — current pick for small sites; pure keyword, build-time index, no service to call.
- Algolia — current pick for large corpora; hosted, typo-tolerant, no vector retrieval on the free tier.
- Meilisearch Cloud — comparable hosted full-text; hybrid search on paid tiers only.
- Self-hosted Typesense — vector + keyword, but self-host fights the family’s no-self-host posture.
Swap cost
Medium — Orama’s @orama/orama client and indexing pipeline
differ from Pagefind’s. The swap involves: (a) replacing the
<Search> wrapper in
re-running the index build at deploy time, (c) wiring an embeddings step into the build for vector retrieval (optional).
Why deferred (not active, not rejected)
- Active is premature — the family’s current search UX is serving users. Adding vector search before there’s a concrete feature that needs it spends complexity for no gain.
- Rejected is wrong — Orama is genuinely the best free hybrid-search candidate the family has surveyed; it would be the swap target the moment the use-case lands.
- Deferred is honest — documents the candidate, the trigger conditions, and the swap surface so the next agent doesn’t re-derive the analysis when the use-case appears.
Cross-refs
- search services index
- Pagefind — current pick for small sites
- Algolia — current pick for large corpora
- Never hit quotas rule
- No card-on-file rule