- BFSIBanking, Financial Services & Insurance
- IT & ITESIT services, BPM, GCCs & SaaS
- Retail & e-commerceStore chains, D2C brands & marketplaces
- HealthcareComing soon
- ManufacturingComing soon
Elasticsearch
The engine under the whole Elastic Stack — Elasticsearch is a distributed search and analytics engine, open source again under AGPL, and increasingly the vector database behind AI retrieval.
Buy through TechBag
Same software. Better outcome — at a lower cost.
How it’s rated
Full scoreboard ↓Where your index is stored — answered by deployment
Self-managed
Stored wherever you run it
On your own servers or your own cloud account, including on-premises. You choose the storage location outright.
Elastic Cloud
Mumbai & Pune regions listed
Hosted lists AWS and GCP Mumbai and Azure Pune; Serverless lists GCP Mumbai. Confirm backups, processing and support access in writing.
For product search and RAG, the index often holds customer records or internal documents. Self-managed keeps storage entirely in your hands; Elastic Cloud offers an Indian region to choose. These are storage regions, not a processing commitment. See the Databases & Data Tools guide for how other engines compare.
Quick answer
The Elastic platform family
This page covers Elasticsearch — the search, analytics and vector engine. The other pillars:
Quick facts
30-second orientation- Product
- Elasticsearch (search, analytics & vector engine)
- Vendor
- Elastic (NYSE: ESTC)
- Licence
- AGPLv3, ELv2 or SSPL — AGPL added in 2024
- Deployment
- Self-managed · Elastic Cloud Hosted · Serverless
- Query languages
- Query DSL and ES|QL (GA since 8.14)
- Vector & semantic
- Dense + sparse vectors; semantic_text (GA 9.0)
- Priced on
- Nodes/RAM · provisioned resources · VCUs
- India regions
- Hosted: AWS & GCP Mumbai, Azure Pune
- Analyst standing
- Leader, Forrester Wave Cognitive Search Q4 2025
- In India via
- TechBag — tier scoping, GST invoicing
Understand the search engine before you buy it
Most product pages skip this. We start here — so you buy a capability, not a buzzword.
What is it?
A distributed search and analytics engine built on Apache Lucene — full-text, vector and hybrid search plus aggregations on one index, behind a REST API.
A bolt-on vector database vs one Elasticsearch index — the honest table
What keeping text and vectors in one engine replaces, dimension by dimension.
| Dimension | Search engine + separate vector DB | Elasticsearch |
|---|---|---|
| Stores | Search index plus a vector database | One index holds text and vectors |
| Sync | Pipelines keep two copies aligned | One write, both representations |
| Hybrid query | Merge result lists in app code | Retrievers fuse results server-side |
| Embedding | Your own embedding service | semantic_text via inference endpoints |
| Filters & facets | Split across two systems | Filters, aggregations and kNN together |
| Licence | Varies per component | AGPLv3, ELv2 or SSPL |
| Honest caveat | — | One bigger system to operate well |
| Best fit | — | Search plus AI retrieval at scale |
The one-index answer — if you only need vectors next to relational data, pgvector may be simpler (TechBag will say so).
The five pieces of the platform
Vendors love diagrams; buyers need to know what they’re actually operating. Here’s the whole platform, demystified.
Apache Lucene
Every shard is a Lucene index: inverted indexes for text, HNSW graphs for vectors, columnar doc values for aggregations. One document, searchable all three ways.
Shards & nodes
Indices split into primary and replica shards spread across nodes, so capacity and resilience grow by adding nodes. Shard count and size are what most often go wrong.
Mappings & analyzers
Mappings declare each field’s type — text, keyword, dense_vector, semantic_text — and analyzers decide how text is tokenised. Relevance is won or lost here.
Inference endpoints
Endpoints connect the cluster to ELSER, E5 or third-party models so semantic_text fields embed at index time — via the Elastic Inference Service or your own ML nodes.
Deployment model
Self-managed, licensed on nodes and RAM; Elastic Cloud Hosted, priced on provisioned resources; or Serverless, billed on ingest, search and ML compute units plus storage.
One Lucene-based index answers keyword, vector and analytical queries — which is why the same engine sits under search apps and log platforms alike.
Twelve capabilities. Index, retrieve, run.
Elasticsearch stores each document once and makes it searchable by keyword, by meaning and by aggregate. Here is what that looks like in practice, grouped by where it happens.
Mappings and analyzers
Typed fields plus language analyzers, synonyms and custom tokenisers — decided up front, and the base of every relevance result.
semantic_text fields
Declare a field semantic and Elasticsearch chunks and embeds it at index time through an inference endpoint. GA since 9.0.
Dense and sparse vectors
dense_vector for embeddings from any model; sparse vectors for learned sparse models like ELSER, which Elastic recommends for English.
Ingest pipelines
Processors that parse, enrich and transform documents on the way in, so less data preparation lives in your application code.
Full-text search
Match, phrase, fuzzy and multi-field queries scored with BM25 — still the baseline most hybrid systems are measured against.
Approximate kNN with BBQ
HNSW vector search; from 9.1, Better Binary Quantization is the default for 384+ dimensions, cutting vector memory about 32x.
Hybrid retrieval and reranking
Retrievers fuse lexical and vector results, for example with reciprocal rank fusion, and can rerank through an inference endpoint.
ES|QL piped queries
A piped query language, GA since 8.14, for filtering, transforming and aggregating data in one readable statement.
Aggregations and analytics
Terms, histograms, percentiles and more, computed over the same index you search — the reason it doubles as a log store.
Index lifecycle and data tiers
Policies roll indices over and move them between hot, warm, cold and frozen tiers as data ages, with no application changes.
Snapshots and searchable snapshots
Back up to object storage and, on the Enterprise tier, query snapshots from the frozen tier instead of restoring them first.
Roles, API keys and restrictions
Role-based access and API keys, with field- and document-level restrictions in paid tiers, so one cluster can serve several teams.
Watch Elasticsearch in action
The data model, vector search and embeddings, explained by Elastic.
How Elasticsearch Works: Documents, JSON & Index Explained
The data model, explained by Elastic.
What Is Vector Search? Vector vs Semantic Search
Why the engine is now an AI retrieval layer.
What Are Vector Embeddings? (Explained in 2 Minutes)
The two-minute version of the AI use case.
Want a live, India-context walkthrough for your environment?
Book a guided demo →One engine for text, vectors and analytics.
Here’s what genuinely sets Elasticsearch apart (and where OpenSearch, pgvector or Pinecone may fit better).
One index for keywords, vectors and analytics
Many AI retrieval stacks bolt a vector database onto a search engine and a warehouse. Elasticsearch keeps BM25 text, dense and sparse vectors and aggregations on the same documents, so hybrid queries, filters and facets run in one place with no copies to sync.
semantic_text takes the plumbing out of RAG
Chunking, embedding and storing vectors is usually glue code you maintain. A semantic_text field does it at index time through an inference endpoint — ELSER, E5 or a provider such as OpenAI or Cohere — so choosing a model becomes a mapping decision, not a service.
Open source again, with a choice of licence
Since August 2024 Elasticsearch can be used under AGPLv3 as well as ELv2 and SSPL — an OSI-approved option that answers the concern which pushed many teams to OpenSearch in 2021. Paid tiers still gate features such as ML-based semantic search.
Three ways to run the same engine
Self-managed on your hardware, licensed on nodes and RAM; Elastic Cloud Hosted, with Mumbai regions on AWS and GCP and Pune on Azure; or Serverless, where you stop sizing clusters and pay for compute units and storage. The APIs you build against stay the same.
The honest caveat — it rewards expertise
Elasticsearch will do almost anything, which means it will also let you do it badly. Oversized shards, mapping explosions, untuned analyzers and deferred upgrades are the usual failures. Budget real engineering time for it, or pay Elastic Cloud to carry the operations.
The honest positioning
Choose it when you need text relevance, vectors and analytics together at scale, or you already run it. If you only need vectors beside relational data, pgvector may be enough; if policy demands Apache 2.0, OpenSearch; for a hands-off vector-only service, Pinecone.
The numbers behind the platform
What your Elasticsearch rollout looks like
Audit what you already run
Many teams already run a cluster for logs or site search. Start with its version, licence and shard layout — TechBag does this free.
Choose deployment and tier
Self-managed, Hosted or Serverless sets cost and storage region; on Hosted, ML semantic search starts at the Platinum tier.
Model mappings, then index
Design mappings, analyzers and semantic_text fields before bulk indexing — changing a field type later means a reindex.
Tune relevance and scale
Test BM25, vector and hybrid on real queries, then size shards for growth. TechBag supports and advises, in INR/GST.
The review scoreboard
Modelled on Gartner Peer Insights structure. *Counts and breakdowns are illustrative pending verified review collection.
Quick poll — what’s driving your evaluation?
Talk to an advisor“Moving our catalogue search to hybrid retrieval cut zero-result queries without throwing away years of BM25 tuning.”
“semantic_text removed a whole embedding microservice from our RAG pipeline. The mapping change took an afternoon; reindexing took longer.”
“We had drifted to OpenSearch after the 2021 licence change. The AGPL option put Elasticsearch back on our approved list.”
“BBQ made our vector index affordable in RAM. Validate recall on your own queries before trusting anyone’s benchmark.”
“Oversharding was our biggest self-inflicted wound. Get shard sizing right early or plan for a painful reindex later.”
“ES|QL let our analysts query without learning the JSON DSL, and usage spread well beyond the platform team.”
“A Mumbai region on Elastic Cloud settled where our index is stored; we still confirmed support-access terms in the contract.”
“We tried pgvector first. Fine at small scale, but facets, typo tolerance and relevance tuning brought us back here.”
Where everyone sits — the grids
Analyst firms bury this view behind paywalls, and G2 retired its Grid. So here’s TechBag’s synthesis of the search and vector-engine market — tap any vendor to see why it sits where it does.
TechBag Search & Vector Engine Grid
Execution strength vs product vision — the classic market map, minus the paywall.
Forrester Leader, Cognitive Search Platforms Q4 2025 — breadth plus hybrid retrieval.
Retrieval Breadth × Deployment Control
The grid nobody publishes — how much retrieval one engine covers vs how freely you choose where it runs.
Text, vectors and analytics, runnable anywhere — the corner it owns.
Positions are TechBag’s illustrative synthesis of public review-platform data and vendor documentation — not a reproduction of any analyst graphic. Verify before relying on it.
Elasticsearch vs the field
The search and vector field — honest lanes; the edge is keywords, vectors and analytics on one index you can run anywhere. Strict Apache 2.0? OpenSearch. Vectors beside Postgres? pgvector. We say so.
| Dimension | Elasticsearch | OpenSearch | Apache Solr | pgvector (PostgreSQL) | Pinecone |
|---|---|---|---|---|---|
| Position | Search, analytics and vector engine | Apache-2.0 fork of Elasticsearch | Long-standing Lucene search server | Vector search inside PostgreSQL | Managed vector database |
| Licence | AGPLv3, ELv2 or SSPL | Apache 2.0 | Apache 2.0 | Open source (PostgreSQL licence) | Proprietary service |
| Deployment | Self-managed, Hosted or Serverless | Self-managed or Amazon OpenSearch Service | Mostly self-managed | Anywhere PostgreSQL runs | Managed cloud service |
| Full-text relevance | BM25, analyzers, synonyms, fuzzy | BM25 from shared Lucene roots | BM25, mature faceting | PostgreSQL full-text, basic ranking | Vector-first; limited text search |
| Vector & hybrid | HNSW, BBQ default, hybrid retrievers | k-NN plugin and hybrid search | Dense-vector kNN; less AI tooling | HNSW and IVFFlat indexes | Purpose-built vector search |
| Analytics & aggregations | Aggregations plus ES|QL | Aggregations, SQL and PPL | Facets and stats | SQL over relational data | Not an analytics engine |
| Built-in embedding | semantic_text + inference endpoints | Neural search with ML connectors | Mostly bring your own embeddings | Bring your own embeddings | Hosted embedding models |
| Operational burden | Yours self-managed; none on Serverless | Yours, or AWS-managed | Yours to run | Low if you already run Postgres | Fully managed |
| India storage region | Self-managed, or Hosted in Mumbai/Pune | Self-managed, or AWS Mumbai | Wherever you run it | Wherever your Postgres runs | Check the current region list |
| Evaluation route | Free engine or Cloud trial | Free and open source | Free and open source | Free extension | Free starter tier, then usage |
| Best fit | Text, vectors and analytics together | Apache-2.0 mandates; AWS estates | Existing Solr estates | Vectors beside relational data | Hands-off, vector-only workloads |
Does Elasticsearch fit you?
Honest fit signals — because the fastest way to lose your trust is to pretend one product wins every scenario.
Choose Elasticsearch if…
- ✓You need full-text relevance, vectors and aggregations on the same data
- ✓Your team already runs Elasticsearch or the wider Elastic Stack
- ✓You want one engine you can run self-managed, hosted or serverless
- ✓You want embeddings handled at index time with semantic_text
Choose OpenSearch if…
- ✓Policy requires an Apache-2.0 engine, or you are standardised on Amazon OpenSearch Service
Choose Apache Solr if…
- ✓You already run Solr well and your needs are mainly classic text search and faceting
Choose pgvector if…
- ✓Your vectors sit beside relational data in PostgreSQL and the scale is modest
Choose Pinecone if…
- ✓You want a managed, vector-only service and have no appetite for running a cluster
Elasticsearch is one of 25 databases & data tools products TechBag carries. The Databases & Data Tools guide narrows them to a shortlist and shows the reasoning. →
Self-managed, or managed?
Elasticsearch is not priced per seat: self-managed subscriptions are licensed on nodes and RAM, and Elastic Cloud Hosted on provisioned resources — so the question that decides your bill is whether running the cluster yourself costs less than letting Elastic run it. The engine is free; your engineers’ time on shard sizing, mappings, capacity and upgrades is not. Drag both sliders. Illustrative TechBag assumptions (not Elastic list prices), at ₹12,000 per fully-loaded engineer-day and ~₹84/USD.
Engineer-days means real time on shard sizing, mappings and reindexing, capacity planning and version upgrades — not application development. If you put 0, you are assuming the cluster runs itself, which is the assumption this calculator exists to question. Illustrative: your TechBag quote models your real tier and resources.
Three ways to run Elasticsearch
Each deployment has its own pricing basis: self-managed subscriptions are licensed on nodes and RAM, Elastic Cloud Hosted on provisioned resources with a subscription tier, and Serverless on usage — compute units for ingest, search and ML plus storage retained. The engine itself is free, so the real questions are which paid features you need and who runs the cluster. Figures below are Elastic’s published “as low as” rates in USD. TechBag scopes tier and deployment and quotes in INR with GST.
Self-managed
Best for full control and on-premises
- Paid subscriptions add features and support, by quote
- Licensed on the number of nodes and RAM used
- You operate the cluster — price your engineers’ time
Elastic Cloud Hosted
Best if you want it managed
- As low as $99 Standard, $114 Gold, $131 Platinum, $184 Enterprise a month
- Starting prices assume 120 GB storage across 2 zones; scales with resources
- ML-based semantic search starts at the Platinum tier
Serverless
Best for variable load
- Ingest VCUs as low as $0.14/hr; ML VCUs as low as $0.07/hr
- Storage as low as $0.047 per GB retained a month; no cluster to size
- Search Power setting trades latency against cost
Buy it for less — TechBag pricing beats list
Whatever the list prices above, TechBag negotiates a significantly better deal — with GST-compliant INR invoicing and local support. Ask us for your discounted quote.
Get an India-ready quote
Tell us your requirements and current tools — we’ll model it against what you spend today.
The 8 questions to ask before you choose a search engine
Take this into your next vendor call — including ours.
Is this full-text search, vector/RAG retrieval, analytics — or all three? The more you need together, the stronger Elasticsearch’s case.
Self-managed, Elastic Cloud Hosted or Serverless? It fixes your pricing basis — nodes and RAM, resources, or VCUs — and your storage region.
If managed, which region stores the index? Hosted lists AWS and GCP Mumbai and Azure Pune; Serverless lists GCP Mumbai.
Will your legal team accept AGPLv3, ELv2 or SSPL for how you distribute or host the software?
Which paid features do you need — ML semantic search, searchable snapshots, SSO? That decides Platinum versus Enterprise.
Which model embeds your content — ELSER (English), E5 (multilingual) or a third-party provider — and who pays to reindex when it changes?
Who owns shard sizing, mappings, upgrades and capacity? Price that time, or price Elastic Cloud to carry it.
Elastic prices in USD — have you modelled the subscription and resources in INR with GST?
Questions buyers ask
Ready to evaluate Elasticsearch?
Find out what you already run, work out which tier and deployment you actually need, or get an honest Elasticsearch vs OpenSearch vs pgvector comparison — in INR/GST.
Stats, ratings, review counts and pricing are illustrative and sourced from public materials; verify before purchase.