DOC-6854 Add redis-rb (Ruby) vector search docs: vecsearch page + ruby_home_query_vec [PARKED]

[andy-stark-redis]

Preemptive docs for redis-rb's Query Engine vector search: a new "Index and query vectors" page for Ruby (content/develop/clients/ruby/vecsearch.md) and a run-verified ruby_home_query_vec example set. Follow-up to the parked Query Engine docs (DOC-6843 / #3641), which deliberately deferred vector search pending an embedding-library decision — now resolved by reusing the semantic-cache Ruby choice (informers running the ONNX all-MiniLM-L6-v2 encoder locally).

⛔ Do not merge yet

The Query Engine landed upstream in redis-rb#1356 (merged) but is not yet in a released gem — the latest is v5.4.1 (July 2025), which predates it. Merging now would document APIs (Redis::Commands::Search::*, create_index, vector_field, KNN search) that gem install redis cannot provide. The page carries a preview bannerText and this PR is labelled parked + do not merge yet.

Park manifest

Ticket: DOC-6854
Parked at: 2026-07-17
Trigger to pick up: redis-rb#1356 shipped in a released redis gem (a version later than v5.4.1 whose lib/redis/commands/modules/search is present)
Labels: parked, do not merge yet

Pinned sources (state observed at park time)

Source	State at park	Re-fetch
redis-rb#1356 — Query Engine support	state: closed, merged: true, head SHA 659d885199d09b997172d753a2f67c39ee23769e, base master, milestone: none, updated 2026-07-14 (squash-merged to master as 71706ec)	gh api repos/redis/redis-rb/pulls/1356 --jq '{state, merged, head_sha: .head.sha, base: .base.ref, milestone: .milestone.title, updated_at}'
redis-rb released gem	latest released v5.4.1 (July 2025) — predates the Query Engine; no release yet contains #1356	gh api repos/redis/redis-rb/tags --jq '.[0:5] | map(.name)' (and check rubygems.org/gems/redis/versions)
informers gem (embedding lib)	verified working at 1.3.0 (+ onnxruntime 0.11.3, tokenizers 0.7.0) on Ruby 4.0.6; ONNX all-MiniLM-L6-v2, 384-dim — already released, NOT a merge blocker (listed for provenance)	gem list informers onnxruntime tokenizers

Observed shape the page assumes — confidence: MEDIUM

The example was run-verified on 2026-07-17 against the local redis-rb fork (master, PR #1356 merged, head 71706ec) on Redis 8 with informers 1.3.0: ruby -I <fork>/lib home_query_vec.rb exited 0, all REMOVE-block asserts passed, and the hash and JSON paths returned identical results. Confidence is MEDIUM rather than HIGH only because it was verified against fork master, not a released gem, so the surface could still shift before release. Key assumptions:

• Schema DSL Search::Schema.build { text_field; tag_field; vector_field 'embedding', 'HNSW', type: 'FLOAT32', dim: 384, distance_metric: 'L2' }; vector_field extracts :as as a field-level option (verified in lib/redis/commands/modules/search/schema.rb), so the JSON path alias vector_field '$.embedding', 'HNSW', as: 'embedding', … works.
• KNN query via index.search('*=>[KNN 3 @embedding $vec AS vector_distance]', sort_by:, return_fields:, params: { 'vec' => bytes }); sort_by defaults to ascending (nearest-first). Document#[] exposes doc['vector_distance'].
• Search::Index strips the index key prefix from returned doc IDs — doc.id is 0, not doc:0 (unlike redis-py). The page/example call this out.
• Vector encoding: hash fields store raw little-endian float32 (Array#pack('e*')); JSON stores the embedding as a plain Array<Float>, but the query parameter is packed bytes either way.
• Embeddings from informers (model.(text, pooling: 'mean', normalize: true), flat 384-element Array) reproduce the canonical L2 distances to ~7 significant figures — 0.114169895649 / 0.610845208168 / 1.48624789715. Normalisation (unit vectors) is load-bearing for those distances (see commit Constraint: trailers).

Re-check checklist

• (highest risk) Re-run local_examples/client-specific/ruby/home_query_vec.rb against the released gem on Redis 8.x; confirm all steps + asserts still pass and the API shape above is unchanged. (Note: the Redis TCP connection needs the sandbox OFF — Operation not permitted on localhost otherwise.)
• Confirm the run-verified distances in both expected-output blocks of vecsearch.md still hold on the released gem; update if they shift.
• Add a "requires redis-rb vX.Y+" note to vecsearch.md once the release version is known.
• Remove the preview bannerText from vecsearch.md once released.
• Confirm examples.json regenerates the Ruby tab from ruby_home_query_vec in a full CI build (not run locally).
• When reconciling, preserve the commit Constraint: trailers — keep normalize on for every encode, and keep the embedding model/library aligned with the semantic-cache Ruby example — or stored vectors stop being comparable across examples.
• No command-api-mapping work in this PR (the FT.* family is covered by DOC-6843 / DOC-6843 Add redis-rb (Ruby) Query Engine docs: queryjson [PARKED] #3641).

On unpark, then

When the trigger fires, run /unpark <this PR number>: it diffs each pinned source against the snapshot above, walks this checklist, then takes the PR through the normal /reflect → /finalize pipeline to merge. The do not merge yet guard holds until /finalize completes.

🤖 Generated with Claude Code

---

Note

Low Risk
Documentation and example-only changes with no runtime or production code; main risk is documenting unreleased redis-rb APIs, which the page banner and park manifest already flag.

Overview
Adds a new Ruby client doc page Index and query vectors (vecsearch.md) that walks through indexing and KNN-searching text embeddings with redis-rb’s Query Engine, using the informers gem and all-MiniLM-L6-v2 locally.

The page is wired to a new run-verified example set ruby_home_query_vec (home_query_vec.rb): hash flow with to_bytes / FLOAT32 packing, then a JSON section covering path aliases, array vs binary storage, and packed query params. It also documents redis-rb behaviors such as default query dialect 2, stripped key prefixes on doc IDs, and a preview banner noting Query Engine APIs are not in a released gem yet.

^{Reviewed by Cursor Bugbot for commit 8aedcc0. Bugbot is set up for automated code reviews on this repo. Configure here.}

[github-actions]

DOC-6854

[github-actions]

Staging links:
https://redis.io/docs/staging/DOC-6854-redis-rb-vecsearch-docs/develop/clients/ruby/vecsearch/