RAG platforms compared (2026)

We read the official sites, docs, pricing pages and GitHub repositories of 14 RAG platforms and vector databases on 2026-10-31. Amazon Bedrock Knowledge Bases, Azure AI Search, Elastic and Zilliz Cloud / Milvus list an Australian region for the product, and Amazon Bedrock Knowledge Bases, Azure AI Search, Elastic, Google Vertex AI RAG Engine, LlamaCloud / LlamaParse, Pinecone, Progress Agentic RAG (Nuclia) and Zilliz Cloud / Milvus list an EU region. "Not stated" means we found no statement on those pages, not that the feature is missing.

Last checked 2026-10-31. Tools are listed alphabetically. This is not a ranking and we did not test the tools ourselves. Enterprise prices and customer counts are not included because we could not source them.

Comparison table

Tool	Type	Open source / license	Self-host	AU region	EU region	Free plan
Amazon Bedrock Knowledge Bases	cloud RAG service	Not stated	Not stated	Yes	Yes	Not stated
Azure AI Search	cloud RAG service	Not stated	Not stated	Yes	Yes	Free tier on the pricing page: 50 MB storage, up to 3 indexes. Semantic ranker is not available on the dedicated Free tier (checked 2026-10-31)
Contextual AI	managed RAG (ingest + retrieval + generation)	Not stated	Yes	Not stated	Not stated	On-demand plan is pay-as-you-go with "Claim your $25 in free credits" (checked 2026-10-31)
Elastic	search engine with vector search	Yes: AGPLv3, SSPL 1.0 or ELv2 (your choice)	Yes	Yes	Yes	Free trial offered ("Start free trial" on the pricing page); trial limits are not stated there (checked 2026-10-31)
Google Vertex AI RAG Engine	cloud RAG service	Not stated	Not stated	Not listed	Yes	Not stated
LlamaCloud / LlamaParse	managed RAG (ingest + retrieval + generation)	Not stated	Yes	Not listed	Yes	Free plan, $0/month: 10K credits, 5 concurrent parse jobs, 5 indexes, 50 files per index, file upload only (checked 2026-10-31)
MongoDB Atlas Vector Search	vector database	No: SSPL v1 (MongoDB server)	Yes	Not stated	Not stated	Free M0 cluster, "Free forever": 512 MB storage, shared RAM and vCPU. The pages we read do not say whether Vector Search runs on M0 (checked 2026-10-31)
Pinecone	vector database	Not stated	Yes	Not listed	Yes	Starter (free): AWS us-east-1 only, up to 5 indexes, 2 GB storage, 2M write units and 1M read units per month; Assistant on Starter has 1 GB storage and a USA region only (checked 2026-10-31)
Progress Agentic RAG (Nuclia)	managed RAG (ingest + retrieval + generation)	NucliaDB (database only): AGPLv3	Not stated	Not listed	Yes	No permanent free tier. Starter plan is "free for 14 days", then a paid plan is required (checked 2026-10-31)
Qdrant Cloud	vector database	Yes: Apache-2.0	Yes	Not stated	Not stated	Free Tier, "Free forever": 1 node, 0.5 vCPU, 1 GB RAM, 4 GB disk, about 1 million 768-dimension vectors; suspended after 1 week of inactivity (checked 2026-10-31)
Ragie	managed RAG (ingest + retrieval + generation)	Not stated	Not stated	Not stated	Not stated	Developer plan, Free: 1,000 retrievals max, 1,000 pages processed max, 10 GB streaming, 10 hours audio/video (checked 2026-10-31)
Vectara	managed RAG (ingest + retrieval + generation)	Not stated	Yes	Not stated	Not stated	30 Day Free Trial, "All features included for 30 days"; trial usage limits not stated (checked 2026-10-31)
Weaviate Cloud	vector database	Yes: BSD-3-Clause	Yes	Not stated	Not stated	Free, "Forever $0": 1 cluster per user, 100,000 objects, 1 GB memory, 10 GB disk, 1 collection, up to 3 tenants (checked 2026-10-31)
Zilliz Cloud / Milvus	vector database	Yes: Apache-2.0 (Milvus)	Yes	Yes	Yes	Free plan, $0: 5 GB storage, 2.5M vCUs per month, up to 5 collections (checked 2026-10-31)

"Yes" for a region means the vendor's own docs list that region for this product, not just for its parent cloud. "Not listed" means the product's region list has no such region. "Not stated" means we found no region list for the product. "Yes" for self-host includes documented private, VPC or bring-your-own-cloud deployments, which the cards describe. See each card for the wording and sources.

Tool by tool

Amazon Bedrock Knowledge Bases

AWS service that converts your data into vector embeddings stored in a vector database and queries them with Retrieve and RetrieveAndGenerate API calls.

Type: cloud RAG service

Suits: Teams already on AWS that want retrieval inside their AWS account and region.

Regions: AWS's Knowledge Bases supported-regions page and the Bedrock agent endpoint list include Asia Pacific (Sydney) ap-southeast-2 and Europe (Frankfurt), (Ireland), (London) and (Paris). London is not in the EU. Features such as reranking and parsing have their own region lists, so check each feature you need.

Notes: The docs warn that with cross-Region inference your data can be shared across Regions. A free plan is not stated on the pages we read.

Sources: docs.aws.amazon.com/bedrock/latest/userguide/kn, docs.aws.amazon.com/general/latest/gr/bedrock.h

Azure AI Search

Microsoft describes it as a fully managed, cloud-hosted service with full-text, vector, hybrid and multimodal queries, plus agentic retrieval for RAG.

Type: cloud RAG service

Suits: Teams on Azure or Microsoft 365 data sources that need a documented Australian or EU region.

Regions: Microsoft's region page lists Australia East and Australia Southeast. In Europe it lists, among others, France Central, Germany West Central, North Europe, West Europe, Sweden Central and Poland Central. Features differ by region (for example agentic retrieval and semantic ranker), and the page marks some regions as in high demand for new services.

Notes: Managed service only on the pages we read. Check the region table for the specific feature you need.

Sources: learn.microsoft.com/en-us/azure/search/search-w, learn.microsoft.com/en-us/azure/search/search-r, azure.microsoft.com/en-us/pricing/details/searc

Contextual AI

A platform and component APIs for building retrieval-based AI agents, priced by usage.

Type: managed RAG (ingest + retrieval + generation)

Suits: Teams that want a managed retrieval platform and are willing to ask the vendor about regions and deployment.

Regions: The pricing page lists no regions.

Notes: The pricing page lists Contextual SaaS and Customer VPC deployment. It does not say which plan includes the VPC option. Enterprise pricing is custom.

Sources: contextual.ai/pricing

Elastic

Elasticsearch on Elastic Cloud. Elastic's GitHub page calls it a vector database optimized for speed and relevance on production-scale workloads.

Type: search engine with vector search

Suits: Teams that already run Elasticsearch or need keyword and vector search together.

Regions: Elastic Cloud Serverless regions page lists AWS ap-southeast-2 (Sydney) and Azure australiaeast (Australia East). EU regions include AWS eu-central-1 (Frankfurt) and eu-west-1 (Ireland), Azure germanywestcentral and northeurope, and GCP europe-west3 and europe-west4. Serverless has no Australian GCP region. We did not check the separate Elastic Cloud Hosted region list.

Notes: Elastic's licensing FAQ says the source is available under AGPLv3 (OSI approved), SSPL 1.0 or ELv2, and that official releases continue under the Elastic License. Docker start-local is for local development only, per the README. The pricing page lists self-managed Elasticsearch as an option.

Sources: elastic.co/pricing/cloud-hosted, elastic.co/docs/deploy-manage/deploy/e, github.com/elastic/elasticsearch, elastic.co/pricing/faq/licensing

Google Vertex AI RAG Engine

Google describes RAG Engine as a component of its Gemini Enterprise Agent Platform that facilitates retrieval-augmented generation.

Type: cloud RAG service

Suits: Teams on Google Cloud that can accept the documented region list and the data residency limitation.

Regions: The RAG Engine region table does not list australia-southeast1. In Europe it lists europe-west3 (Frankfurt) and europe-west4 (Eemshaven) as GA, and several others (such as europe-west9 Paris) as Preview or allowlist. The page also says: "Data residency and AXT security controls aren't supported."

Notes: Read the data residency line with your compliance team before choosing it for regulated data. Some regions need allowlisting.

Sources: docs.cloud.google.com/vertex-ai/generative-ai/doc

LlamaCloud / LlamaParse

LlamaParse, part of LlamaCloud, says it transforms messy documents into AI-ready data, with Parse, Extract and Index offerings.

Type: managed RAG (ingest + retrieval + generation)

Suits: Teams whose main problem is parsing complex documents before retrieval.

Regions: LlamaCloud docs list two regions: North America (AWS us-east-1) and Europe (AWS eu-central-1). Australia is not listed.

Notes: Private VPC deployment is described for enterprises ("SaaS or Hybrid cloud deployment" on the Enterprise plan). It is not a self-serve self-hosted product, and the pricing page does not describe a fully on-premises option.

Sources: llamaindex.ai/pricing, llamaindex.ai/llamaparse, developers.llamaindex.ai/python/cloud/general/region

MongoDB Atlas Vector Search

MongoDB says you can build vector search, hybrid search and generative AI apps on the live operational data already in MongoDB.

Type: vector database

Suits: Teams already storing their application data in MongoDB.

Regions: MongoDB's Atlas AWS region list includes ap-southeast-2 (Sydney) and EU regions such as eu-central-1 (Frankfurt), eu-west-1 (Ireland) and eu-west-3 (Paris) for dedicated clusters. The Vector Search pages we read do not state regional availability for Vector Search itself, so we left both fields unknown. Check the Vector Search compatibility page.

Notes: MongoDB says Vector Search is also available in MongoDB Community Edition and Enterprise Advanced for self-managed use. The server repository states releases after October 16, 2018 are under SSPL v1, which is source-available, not open source.

Sources: mongodb.com/products/platform/atlas-vec, mongodb.com/docs/atlas/atlas-vector-sea, mongodb.com/docs/atlas/reference/amazon, mongodb.com/pricing, github.com/mongodb/mongo

Pinecone

Managed vector database with serverless indexes. Pinecone Assistant is a separate service for building chat and agent applications.

Type: vector database

Suits: Teams that want a managed vector database and can use an EU region, or the Assistant for a quicker chat app.

Regions: Serverless index regions in the docs: AWS us-east-1, us-west-2, eu-west-1 (Ireland), eu-central-1 (Frankfurt), ap-southeast-1 (Singapore); GCP us-central1, europe-west4 (Netherlands); Azure eastus2. No Australian region is listed. Pinecone Assistant takes region "us" (default) or "eu" and the region cannot be changed later.

Notes: Self-host is true only as Bring Your Own Cloud: Pinecone runs in your cloud account, with sizing and pricing through sales. It is not an open-source server. The open-source status of the product is not stated.

Sources: pinecone.io/pricing/, docs.pinecone.io/guides/index-data/index-con, docs.pinecone.io/guides/assistant/create-ass

Progress Agentic RAG (Nuclia)

Progress Agentic RAG, formerly Nuclia, is a hosted RAG platform built around NucliaDB, a hybrid search database with vector, full-text and graph indexes.

Type: managed RAG (ingest + retrieval + generation)

Suits: Teams that want an EU-hosted RAG service, or want to keep source data in their own NucliaDB instance.

Regions: The Data Ownership docs say you can choose between two regions: Europe or the United States. No Australian region is listed.

Notes: You can run your own NucliaDB instance, but the docs say processing still happens on the Agentic RAG cloud platform, so this is not the full product on your own infrastructure. The pricing page lists "on-prem options" under Enterprise. NucliaDB is open source under AGPLv3, the platform as a whole is not stated as open source.

Sources: progress.com/agentic-rag/pricing, docs.rag.progress.cloud/docs/management/security/da, github.com/nuclia/nucliadb

Qdrant Cloud

Managed cloud for Qdrant, an open-source vector database and similarity search engine built in Rust.

Type: vector database

Suits: Teams that want an open-source engine they can run themselves, with the option of a managed cluster.

Regions: The docs and pricing page name AWS, Azure and Google Cloud but do not list regions; they point to the Cloud Dashboard region selector, the pricing calculator and the qcloud CLI. We could not confirm an Australian or EU region from the pages we read.

Notes: Docker self-hosting is in the README, which warns the default deployment has no authentication. Qdrant also lists Hybrid Cloud (managed clusters on your infrastructure) and Private Cloud (enterprise, contact sales).

Sources: qdrant.tech/pricing/, qdrant.tech/documentation/cloud/create-, github.com/qdrant/qdrant

Ragie

A RAG platform that handles document, image, audio and video ingestion with hybrid search, reranking, entity extraction, connectors and an MCP server.

Type: managed RAG (ingest + retrieval + generation)

Suits: Small teams that want ingestion, connectors and retrieval as one managed API.

Regions: The pricing page has no region, self-hosting or on-premises details. Enterprise is custom pricing through sales.

Notes: Ask Ragie directly about data region and deployment options before sending regulated data.

Sources: ragie.ai/pricing

Vectara

An API-first agentic platform for building AI agents that retrieve knowledge and give grounded answers.

Type: managed RAG (ingest + retrieval + generation)

Suits: Teams with strict residency needs that can run a private deployment, or that want a managed API.

Regions: The pages we read name no SaaS regions. Vectara's docs describe private deployment (on-premises, cloud VPC, air-gapped, Kubernetes) where processing takes place where your information resides.

Notes: Self-host is true as a private deployment, which the pricing page lists as a separate plan from SaaS. It is not an open-source server. Enterprise pricing is not given here.

Sources: vectara.com/pricing, docs.vectara.com/docs/, docs.vectara.com/docs/deployments/private-de

Weaviate Cloud

Managed cloud for Weaviate, an open-source, cloud-native vector database that stores objects and vectors.

Type: vector database

Suits: Teams that want an open-source database they can self-host or move to a managed cluster later.

Regions: The pricing page gives region counts by plan (Free: AWS, 2 regions; Flex: GCP and AWS, 7; Premium: GCP, AWS and Azure, about 40) but does not name regions. The docs say you pick the region in the create-cluster form. We could not confirm an Australian or EU region from the pages we read.

Notes: The repository lists Docker and Kubernetes as install options. A wl/ folder and enterprise features are under a commercial license.

Sources: weaviate.io/pricing, docs.weaviate.io/cloud/faq, github.com/weaviate/weaviate

Zilliz Cloud / Milvus

Zilliz Cloud is the managed service for Milvus, an open-source vector database.

Type: vector database

Suits: Teams that want an open-source engine with a managed option that lists a Sydney region.

Regions: Zilliz's region page lists AWS ap-southeast-2 (Sydney); on-demand, free and serverless, and dedicated clusters are available there, always-on serving clusters are not. EU regions listed: AWS eu-central-1 (Frankfurt) and eu-west-1 (Ireland); GCP europe-west3 (Frankfurt), europe-west4 (Eemshaven) and europe-west9 (Paris); Azure Germany West Central and North Europe. Azure and GCP have no Australian region.

Notes: The Milvus repository describes Docker standalone and Kubernetes deployment. Zilliz also lists BYOC through sales.

Sources: docs.zilliz.com/docs/cloud-providers-and-re, zilliz.com/pricing, github.com/milvus-io/milvus

How to choose

1. Start from your data residency requirement

If data must stay in Australia, only Amazon Bedrock Knowledge Bases, Azure AI Search, Elastic and Zilliz Cloud / Milvus list an Australian region for the product in the pages we read. For the EU, the list is longer: Amazon Bedrock Knowledge Bases, Azure AI Search, Elastic, Google Vertex AI RAG Engine, LlamaCloud / LlamaParse, Pinecone, Progress Agentic RAG (Nuclia) and Zilliz Cloud / Milvus. A region keeps stored data in that region. Your embedding model, reranker and answer model may run elsewhere, so ask where each step runs. Google Vertex AI RAG Engine is one example: Google's page says data residency controls are not supported.

2. Decide managed RAG or build it yourself

A vector database gives you storage and search and leaves parsing, chunking and answer generation to you. A managed RAG service covers more of that, with less control. Pick the database route if you already have an engineering team and want to change each step. Pick the managed route if you want a working prototype sooner.

3. Use the cloud you already pay for

Amazon Bedrock Knowledge Bases, Azure AI Search and Google Vertex AI RAG Engine sit inside AWS, Azure and Google Cloud. If your data and security reviews are already there, that can shorten procurement. Check the region table for the specific feature you need, because features have separate region lists.

4. Check document parsing needs early

If your sources are scanned PDFs, tables or slides, parsing is a separate problem from storage and search. LlamaCloud / LlamaParse is built around parsing, and Ragie lists audio and video processing. Test your own documents on each free plan before you commit.

5. Check the license if you plan to self-host

Qdrant Cloud, Weaviate Cloud and Zilliz Cloud / Milvus are under permissive licenses per their repositories. MongoDB Atlas Vector Search (SSPL v1 (MongoDB server)) is source-available, with restrictions. Read the license before you build a hosted product on it. Self-hosting also moves region questions to "where do we run it" and adds operations work.

Also see LLM observability tools compared for tracing and evaluating the app once it runs, or the wider AI development overview.

Need someone to build it?

If you would rather hand this to a specialist, see how to hire RAG developers, browse the RAG agencies in our directory, or get matched with an agency. Related lists: agencies for RAG and knowledge bases and agencies for LLM apps, RAG and agents. Our listings do not say which vector database or RAG platform an agency uses, so ask about it in your brief.

Sources

All fetched on 2026-10-31.

Questions

Which RAG platforms can keep data in Australia?

On the vendor documentation we read, Amazon Bedrock Knowledge Bases, Azure AI Search, Elastic and Zilliz Cloud / Milvus list an Australian region for the product. Google Vertex AI RAG Engine, LlamaCloud / LlamaParse, Pinecone and Progress Agentic RAG (Nuclia) publish a region list for the product with no Australian region in it. For the other 6 tools in the table, an Australian region is not stated, which is different from "not available": ask the vendor before you assume either way. Region names and sources are in each card.

Is a vector database the same as a RAG platform?

No. A vector database stores embeddings and returns the closest matches to a query. A managed RAG platform adds the other steps, such as parsing and chunking documents, retrieval and often answer generation. Several tools here are vector databases (MongoDB Atlas Vector Search, Pinecone, Qdrant Cloud, Weaviate Cloud and Zilliz Cloud / Milvus), and others are managed RAG services (Contextual AI, LlamaCloud / LlamaParse, Progress Agentic RAG (Nuclia), Ragie and Vectara). With a vector database you build the pipeline around it yourself.

Which options are open source and self-hostable?

Elastic, Qdrant Cloud, Weaviate Cloud and Zilliz Cloud / Milvus state an open-source or choose-your-license core that you can run yourself. Qdrant Cloud, Weaviate Cloud and Zilliz Cloud / Milvus are under a permissive license (Apache-2.0, BSD-3-Clause and Apache-2.0 (Milvus)) according to the pages we checked. MongoDB Atlas Vector Search (SSPL v1 (MongoDB server)) is source-available, which is not open source, and its self-managed Community Edition is documented. Contextual AI, LlamaCloud / LlamaParse, Pinecone and Vectara document a private, VPC or bring-your-own-cloud deployment, usually through sales, which is not the same as an open-source server.

What does it cost to start?

Azure AI Search, LlamaCloud / LlamaParse, MongoDB Atlas Vector Search, Pinecone, Qdrant Cloud, Ragie, Weaviate Cloud and Zilliz Cloud / Milvus state a free plan or free tier on their pricing pages. Contextual AI, Elastic, Progress Agentic RAG (Nuclia) and Vectara state a time-limited trial or free credits instead. Amazon Bedrock Knowledge Bases and Google Vertex AI RAG Engine did not state a free plan on the pages we read. Limits are in the table and were checked on 2026-10-31. We do not list enterprise pricing because the vendors do not publish it consistently. Confirm on each pricing page.

Should I hire an agency to build RAG?

It depends on who will own it. The platform is the smaller part of the work. Most effort goes into getting documents in, chunking and parsing them well, testing answers against real questions and handling permissions. If you have engineers with time for that, a managed service or a vector database can be built in-house. If you do not, a specialist can scope it with you. Our directory lists agencies that name RAG work on their own websites.

Get matched with 3 vetted agencies

Describe the project. We read every brief and introduce you to three agencies from this index that fit it, usually within 2 working days. Free for buyers; agencies can't pay to be picked.