- Starting price
- Free
- Free tier
- Yes
- Platforms
- Web · public REST endpoint · self-hosted source
- Developer
- Tim / Antioch Tech
- Launched
- 2024
- Updated
- Aug 9, 2026
The verdict
Bible Vector Search is a fast, free semantic lookup for the Bible in Basic English and a clean demonstration of what embeddings can add to Bible discovery. It can surface the verse you mean when your remembered wording is wrong, but it provides no confidence score, context, alternative translation, keyword mode or “no good match” state. Treat every result as a lead to open in a complete Bible, and treat the undocumented public endpoint as a convenience rather than production infrastructure.
Try Bible Vector Search ↗Opens bible-search.antioch.tech
Bible Vector Search is a small web application from Antioch Tech that searches Scripture by semantic similarity instead of requiring the exact words in the verse. Enter an idea such as “God is love,” “love your enemies” or “fear and anxiety,” and it returns five individual verses from the Bible in Basic English. The project embeds Bible verses into a mathematical representation, sends the query through a related representation, and asks a vector database which verse records are nearest. That is useful when ordinary full-text search fails because the words in your memory are not the words in the chosen translation.
The distinction from an AI Bible chatbot is important. The public interface does not compose a devotional, explain doctrine or answer a question in prose. It returns existing verse records with book, chapter and verse. The AI-like component is retrieval: an embedding model determines which meanings look close. That avoids fabricated verse wording, but it does not make relevance objective. A similarity engine can rank an opposing idea highly because it shares concepts, can miss an obvious verse, and will return something even when the query is meaningless.
Live checks made that boundary visible. “GodIsLove” correctly ranked 1 John 4:8 first. “Love your enemies” included Luke 6:27, but placed Psalm 25:19—a statement about enemies’ bitter hatred—ahead of it and also returned passages about destroying or hating enemies. “Fear and anxiety” mixed fear of the Lord, fear as distress and persecution. A nonsense string still produced five apparently unrelated verses. The system offers no score, threshold, explanation or empty state that tells a reader when all five matches are weak. Relevance therefore requires reading the surrounding passage and comparing the query’s intended sense.
Developers also get a no-key GET endpoint at `/api/search?verse_query=...`. The current response is a JSON array with five objects containing an internal index, book name and number, chapter, verse, translation identifier and verse text. The official documentation is one example URL and a single sentence. It does not publish a schema, encoding rules, limits, allowed use, versioning, availability, error contract, CORS support, attribution requirement or deprecation policy. Response headers did not advertise cross-origin access during review, so a browser application on another domain should not assume direct client-side calls will work.
The source repository is public under the MIT license and showed 31 stars, two forks, 19 commits and no release tags during research. The last code commit in the visible history was September 2024; a November 2025 change only added a funding file. The live site and API still worked in August 2026, which suggests a stable maintenance state rather than proof of abandonment. Our 4.1 rating is a researched editorial assessment based on the live interface, repeated API calls, public source, official documentation and project pages—not an embedding benchmark, penetration test, theological review of every result or uptime measurement.
✓ The good
- Meaning-based discovery - useful when you remember an idea but not the exact wording, reference or translation vocabulary
- Returns real verse records - the tool retrieves stored Scripture instead of asking a language model to invent quotation text
- Immediate and anonymous web use - no account, email, checkout or onboarding stands between a reader and a search
- Public REST response - developers can inspect the same structured results used by the web interface
- Proper open-source license - the repository includes an MIT license rather than merely exposing readable code
- Simple scope - one search box and five results keep the product understandable and free of generated interpretation
- Bible reference included - every card identifies the book, chapter and verse so a result can be checked in context
- Free to use - there is no paid search tier, metered key or advertised subscription
✗ Watch out
- Always returns candidates - nonsense input still receives five verses, with no “no reliable match” threshold
- No relevance evidence - scores, matched concepts, embedding model and corpus-build version are not exposed
- One translation only - every result uses the Bible in Basic English, whose restricted vocabulary can shape semantic matching
- Context is absent - result cards show one verse without the paragraph, chapter, cross-references or link to a full reader
- API documentation is skeletal - no schema, limits, SLA, error contract, versioning, CORS policy or acceptable-use notice
- Search modes cannot be combined - there is no exact phrase, keyword, reference, book, testament or translation filter
- Maintenance is quiet - visible application code has not changed since September 2024 and there are no tagged releases
- Privacy is unstated - the operator does not explain query logging, IP retention, analytics, infrastructure providers or deletion
Best for
- Readers trying to recover a verse from a remembered theme or paraphrase
- Pastors and teachers building an initial list of passages to verify in context
- Developers prototyping semantic Scripture retrieval with a tiny JSON response
- Students learning how vector similarity differs from keyword search
- Open-source contributors who want an inspectable Next.js search example
Avoid if
- You need a complete Bible reader, multiple translations, commentary or original-language tools
- You need ranked evidence that can be accepted without checking the passage
- Your application requires a supported API, service agreement, fixed quota or documented data-processing terms
- You need client-side cross-origin calls and cannot place a controlled server between your app and the endpoint
- You are searching for an exact word, phrase or reference that a conventional concordance handles deterministically
What Bible Vector Search is
A vector search does not compare only literal tokens. An embedding model converts each verse and the user query into numeric vectors intended to represent semantic relationships; nearest-neighbor search retrieves the vectors that sit closest. That can connect “care when I am worried” with a passage about casting anxiety even if the exact query words never occur.
Bible Vector Search applies that approach at verse granularity to the Bible in Basic English. The web frontend calls its own public route, and that route forwards the query to a separately configured backend. The response contains stored verse text rather than generated wording. It is best described as semantic retrieval, not generative Bible interpretation—even though embeddings are machine-learning technology.
Why a remembered meaning can beat remembered wording—and why polarity matters
Concordances reward exact vocabulary. Search “unconditional love” in a translation that never uses that phrase and a literal engine may return nothing. Semantic search can locate verses about steadfast love, mercy, adoption or God loving the world because the ideas are related in the embedding space. This is especially helpful when a sermon, paraphrase or another translation supplied the words in your memory.
The same compression loses important distinctions. Love and hate of enemies occupy a shared conceptual neighborhood; fear of God and anxious fear share a word while functioning differently; a lament and a promise about suffering may look similar. Verse-sized records also separate claims from qualifications in neighboring sentences. The responsible workflow is query, collect, open context, compare translations, then decide. A vector rank accelerates discovery but cannot settle interpretation.
Five semantic matches: fast discovery without a reliability threshold
The interface offers one search field, a Search button and a vertical result list. Each card contains one BBE verse and a reference chip. A request returns five records, apparently in nearest-first order. There are no visible scores, query refinements, filters, pagination, canonical grouping, result explanations, feedback controls, direct reader links or option to ask for more or fewer matches.
Five is a reasonable brainstorming size, but fixed output becomes misleading when the nearest neighbor is still far away. A strong system should expose normalized similarity, label the embedding and corpus revision, suppress results below a validated threshold, detect reference-shaped input, offer keyword fallback and invite the reader to expand context. Until then, the first card is simply the closest candidate the system found—not “the biblical answer.”
BBE-only corpus: accessible English with a strong translation fingerprint
The Bible in Basic English was designed around a limited English vocabulary and simplified constructions. That makes many sentences approachable and gives the project a distributable English corpus. Every live result identified `ENGLISHBBE`; there is no edition picker or parallel display. Search behavior is therefore shaped both by the source text and by the embedding model built over it.
A limited vocabulary may broaden some conceptual matches while obscuring distinctions expressed differently in another translation. Pastors should not paste BBE text into material labeled NIV, ESV, KJV or another edition. Readers should open the reference in the translation they normally study, read at least the surrounding paragraph and preserve attribution when quoting. Translation comparison is not a cosmetic add-on here; it is the main safeguard against treating one wording as the entire search universe.
REST endpoint and MIT frontend: an inviting prototype with missing production contracts
The endpoint accepts `verse_query` and returns JSON fields for internal index, book name and number, chapter number, translation name, verse number and text. A valid request received HTTP 200. An explicitly empty query returned a structured “Query text is required” detail, while omitting the parameter altogether flowed through the public proxy as the string `null` and still produced five unrelated verses. The open route code reveals simple forwarding plus one-hour Next.js revalidation.
That transparency is useful to learners, but production consumers need more: URL-encoding examples, status codes, response types, maximum length, abuse limits, attribution, caching rights, CORS, change logs and a versioned base path. The frontend repository does not contain the complete embedding backend, so the MIT license should not be mistaken for a reproducible data service. Fork the interface freely; plan a separate retrieval stack if continuity matters.
Pricing
Public web search
Free
The live search page requires no account or payment and returns five BBE verse cards. There is no advertised premium tier, history, saved list or paid relevance feature. Free access is a present product fact, not a promise of perpetual operation or support.
Public JSON endpoint
Free, undocumented
The GET endpoint accepted queries without a key during research. No page defines request quotas, commercial use, rate-limit windows, uptime, data retention, version guarantees or support. Use it for exploration with timeouts and caching; ask the operator before building material dependency on it.
Self-hosted code
MIT code plus your infrastructure
The web repository is MIT-licensed, but the published Next.js route proxies to an `API_URL` environment variable. The vectorization service, vector database contents and deployment recipe are not all packaged as a one-command local stack. Budget for your own corpus, embedding generation, database, hosting, monitoring and data-license review.
There is nothing to buy. The web app has no advertisements, checkout or account wall, and the repository points to optional GitHub sponsorship. That makes Bible Vector Search an excellent low-friction public demonstration and a helpful personal lookup companion.
The API is financially free at the point of access but operationally undefined. “No API key” is not the same as “unlimited,” and a working endpoint is not an SLA. A church site should not make every page render depend on it. Cache lawful results, set short timeouts, monitor response shape and show an honest fallback when retrieval fails.
Self-hosting shifts rather than removes cost. The visible application is a frontend and proxy; teams still need an embedding model, a properly licensed verse corpus, an index, a vector database and a backend compatible with the route. Rebuilding the index is also necessary when the translation, segmentation or model changes.
Data licensing deserves separate treatment from code licensing. MIT clearly covers the repository code. Bible text carries its own publication history and jurisdictional questions, and a future multi-translation fork would need permission for every edition. Put translation name and rights notice beside exported results instead of assuming a code license covers content.
The time cost of verification is part of the product economics. A result can save five minutes of guessing and then cost thirty seconds to open in context; that is a good trade. It is a bad trade only when an organization treats rank one as approved copy and later corrects a misapplied verse in published material.
Where Bible Vector Search falls behind
The result model needs uncertainty. Always returning five items makes the interface look confident even for gibberish. Add a calibrated threshold, “weak match” banner, scores and a genuine no-results state. Validate that calibration against a published set of paraphrase-to-reference and adversarial queries.
Context should be one click away. Every reference needs a chapter link, preceding and following verses, translation label visible on the card, copy-with-citation and a warning when a single verse may be incomplete. Search history and a temporary shortlist would help researchers without requiring an account.
Hybrid search would materially improve precision. Let users switch or combine semantic, exact phrase, all words, any words and direct reference lookup; filter by testament, book and genre; and deduplicate closely related parallel passages. Vector search is a complement to a concordance, not its replacement.
The API documentation is not sufficient for third-party reliance. Publish OpenAPI or JSON Schema, examples with encoded spaces, all error responses, query length, result count controls, rate limits, caching, attribution, allowed use, CORS policy, uptime page, versioning and deprecation windows. Fix the missing-parameter path so `null` is not searched as content.
The retrieval model is a black box. Name the embedding model, vector database, distance metric, verse segmentation, preprocessing, index date and whether searches are logged. Without those facts, no one can reproduce results, evaluate drift or know when a model upgrade changed rankings.
Privacy needs a plain notice. Search phrases can disclose grief, abuse, self-harm, sexuality, doubt or other sensitive concerns. State whether queries, IP addresses and user agents are stored; identify Cloudflare, Vercel and the backend host; give retention periods and contact details; and discourage sending names or confidential pastoral information.
Maintenance signals are thin. The repository has no releases and application code has been quiet since 2024, while the service continues to run. A short current-status document, dependency updates, security contact, backup plan and ownership handoff policy would make the stable experiment much safer to embed.
Bible Vector Search vs. Bible Gateway vs. Bible AI Search
Bible Gateway is the deterministic baseline: many translations, keyword and passage search, full context, audio and familiar citation links. It is better when you know words or a reference and need publishing-grade context. Bible Vector Search is faster when you only remember the concept, but its single BBE corpus and fixed five candidates make it a discovery front end rather than a reader.
Bible AI Search turns natural-language questions into an answer-oriented experience and offers a much larger reader and language surface. That may be more approachable for a person seeking an explanation, but generated or curated answers introduce another editorial layer. Bible Vector Search is refreshingly restrained: it returns verses and stops. The cost of restraint is that users must perform all synthesis and quality control themselves.
OpenBible.info topical pages aggregate community-supplied verse votes for named themes. They are transparent about popularity and often produce broader lists, while Bible Vector Search can handle novel phrasing without waiting for a topic page. The best research workflow can use both: semantic candidates for recall, topic voting for breadth, then a full Bible and responsible commentary for context.
A local concordance or Logos search is stronger for confidential work and reproducibility. It can combine fields, morphology, proximity and owned translations without sending a pastoral query to a small public endpoint. That is more setup and cost. Bible Vector Search wins on zero friction; professional tools win on controls, provenance and continuity.
The bottom line
Bible Vector Search deserves credit for doing less. It does not pretend to be a pastor, prophet or universal Bible answer machine. It applies semantic retrieval to one English translation and returns references a human can examine. On remembered-meaning tasks, that is often exactly the missing tool. The live results also show why semantic proximity cannot be equated with biblical relevance: a command to love enemies can rank beside hatred and destruction because the shared subject dominates the vector, and nonsense still receives polished Scripture cards. We recommend it as a free first-pass discovery aid and an educational API prototype. We do not recommend citing its first result without context, transmitting confidential counseling details, or building a production dependency around the unversioned endpoint. The next release should add uncertainty, hybrid search, context links, model and corpus documentation, privacy terms and a real developer contract. Used within those boundaries, it can turn “I know there is a verse about this” into a short, checkable list faster than literal search alone.
Alternatives to Bible Vector Search
Bible GatewayThe multi-translation reader and exact-search baseline, with whole passages and stable citation links.
Bible HubFree parallel translations, cross-references, interlinear tools and public-domain commentary around a known reference.
Blue Letter BibleA mature free study suite with lexical search, concordance, commentaries and contextual reading.
Bible MCPA remote tool server for exact verse and chapter retrieval once an AI agent already knows the coordinates.
LogosProfessional research software with powerful search fields, owned resources and far stronger provenance at greater complexity.Frequently asked questions
What is Bible Vector Search?
Bible Vector Search is a free Antioch Tech web tool that uses embeddings and vector similarity to retrieve five Bible in Basic English verses related to the meaning of a query. It returns stored Scripture text and references; it does not generate an explanation or prove which result best answers the user.
Is Bible Vector Search accurate?
It can be excellent for verse recall: “GodIsLove” ranked 1 John 4:8 first in our live check. Relevance is inconsistent for ambiguous language. “Love your enemies” also surfaced passages about hating or destroying enemies, and nonsense still produced five verses. Read every candidate in context.
Which Bible translation does it use?
The live product indexes the Bible in Basic English, identified in API results as `ENGLISHBBE`. There is no translation selector. Copy the reference into your usual Bible, compare wording and preserve the correct edition label whenever you quote the returned text.
Is there a free Bible Vector Search API?
Yes. A no-key GET endpoint at `/api/search?verse_query=...` returned five JSON records during research. The public docs provide only one example and no quota, SLA, versioning, CORS, retention or support contract, so it is suitable for experiments rather than an unqualified production dependency.
Is Bible Vector Search open source?
Yes. The public `tim-hub/bible-vector-search` repository includes an MIT license. It covers the Next.js web application and proxy code. The complete separately hosted embedding backend and a reproducible vector corpus are not packaged in that repository, and Bible text rights remain separate from the code license.
Does it use generative AI or hallucinate verses?
Its central function is semantic retrieval rather than prose generation: it returns verse text stored in the indexed corpus. It can still retrieve an irrelevant passage, because embeddings estimate similarity and the interface lacks a reliability threshold. The wording is real BBE text; the match may still be wrong for the intended question.
Is Bible Vector Search private?
No product privacy notice explains whether search text, IP addresses or logs are retained. The live request passes through public hosting to a separate API backend. Avoid names, counseling details or other sensitive information, and use a local or contractually governed search system for confidential ministry work.
Sources & further reading
More AI Bible & Faith Tools
Magisterium AIA chatbot that refuses to answer Catholic questions from the open internet - and instead pulls every claim from 25,000+ Magisterial documents with citations.
BiblosOne of the clearest free demonstrations of meaning-based Bible search, now paired with substantial daily-reading tools—but semantic relevance is not interpretation, and the live service lacks the privacy and methodology pages needed for unqualified trust.
AnorA white-glove intelligence layer that connects a church or nonprofit’s existing email, giving, event, and commerce systems, resolves constituent identities, and answers cross-system questions—powerful for organizations with fragmented data, but costly and governance-heavy because the “complete picture” is sensitive by design.
AskMyChurchA church-trained AI assistant with unusually thoughtful citations, anonymous questions, pastoral boundaries and crisis routing—stronger than a generic chatbot, yet paired with absolute accuracy claims and incomplete retention, vendor and compliance detail.
Bible BotA carefully instructed Christian companion inside ChatGPT, with unusually candid limits and pastoral safety rules—but still a probabilistic AI whose model, privacy, memory, and access belong to OpenAI’s platform.
Church CanvasA ministry-specific AI graphics workspace that makes prompting, variations, resizing, and sermon assets far easier for a non-designer—fast and affordable, but its changing plans and broad content license deserve careful review.