Private cloud
Assistant on Amazon Bedrock
A model served through Amazon Bedrock, Bedrock Knowledge Bases for retrieval over your content (with OpenSearch Serverless if you manage the vector store yourself), a web application you build, IAM Identity Center for sign-in, and a VPC with PrivateLink so the call stays on your network.
- Source
- Editorial recipe, 4 of 6 references fetched and hashed
- Verified
- 20 Aug 2026
- Confidence
- High
01Objective
Run a retrieval-grounded assistant on a frontier model inside your own AWS account, reachable over private networking, with no GPU fleet to operate.
- use caseDocument Q&AAsk questions of a specific document or a small set of them and get an answer with the passage it came from. Narrower than company-wide search: the user already knows which file matters and wants it read carefully.
- use caseInternal company searchOne search box over the documents a company already has — shared drives, wikis, ticket systems, email attachments — answered by a model that quotes the source. The value is finding the right paragraph in a corpus nobody has read end to end, not writing new text.
- use caseKnowledge managementCapturing what the organisation knows — process notes, decisions, answers given once already — and keeping it findable and current. AI helps by drafting entries, spotting duplicates and flagging pages that contradict each other.
- use casePrivate company ChatGPTA chat assistant for staff that behaves like a consumer chatbot but runs under company control: your accounts, your logging, your retention, your choice of model, and optionally your own documents attached.
02Recommended stack
| Role | Component |
|---|---|
| Authentication | IAM Identity Center (or your identity provider) |
| Gateway | VPC with PrivateLink |
| Inference | Amazon Bedrock |
| Inference | Claude Sonnet (or another Bedrock model) |
| Orchestration | Amazon Bedrock Knowledge Bases |
| Ui | Chat application (your own web app) |
| Vector storeoptional | Amazon OpenSearch Serverless |
Architecture and data flow
Components
- Employees (office network) — people
- VPC / PrivateLink — network
- Chat application (your own web app) — application
- IAM Identity Center — identity
- Retrieval layer — retrieval
- Managed vector store (or OpenSearch Serverless) — vector store
- Application store (chats, users) — database
- Amazon Bedrock (managed endpoint) — inference server
- Claude Sonnet (Bedrock model) — model
Connections
- Employees (office network) to VPC / PrivateLink — HTTPS (confidential data)
- VPC / PrivateLink to Chat application (your own web app) — reverse proxy (confidential data)
- Chat application (your own web app) to IAM Identity Center — OIDC sign-in (personal data)
- Chat application (your own web app) to Retrieval layer — question + user groups (confidential data)
- Retrieval layer to Managed vector store (or OpenSearch Serverless) — embedding search (confidential data)
- Retrieval layer to Application store (chats, users) — documents + permissions (confidential data)
- Chat application (your own web app) to Application store (chats, users) — chats, users, settings (personal data)
- Retrieval layer to Amazon Bedrock (managed endpoint) — prompt + retrieved passages (confidential data)
- Amazon Bedrock (managed endpoint) to Claude Sonnet (Bedrock model) — loaded weights
External data transfer · SOME
- confidential content leaves your premises for your own cloud tenancy ("VPC / PrivateLink"). You keep control of the account; the provider is a processor, so a DPA and a documented region apply.
- Everything sits in one AWS account and Region you control; AWS is a processor for the model and the knowledge base, not a recipient of your content for its own purposes.
- Each model provider on Bedrock runs in its own per-Region deployment account — the provider’s terms apply alongside the AWS agreement.
03Suitable for
- Organisation size
- 50–20000 employees
- Data classes
- confidential, personal
- Constraints
- an AWS account with a landing zone, an identity provider and a signed agreement already in place; the model you want granted in the region your residency rule requires; a team, or an implementation partner, who can build a web application
- Industries
- Financial services, Insurance, Professional services, Technology, Retail
- Jurisdictions
- any
04Hardware
No hardware profile was sized for this answer.
Indicative costUSD · one-off plus monthly
- Software
- No third-party licence: the application is yours and the managed services bill on usage. The model bill sits on the model line.
- US$0
- Model usage
- Not estimated here: priced per token — see the cost section, which multiplies a fetched price by the usage band read from your brief. Nothing in this seed asserts a price.
- Not estimated
- Knowledge base, vector store, hosting and network
- Not estimated: the knowledge base, OpenSearch Serverless (if used), application compute, PrivateLink and egress are priced per configuration and Region, and no price list was fetched. Read yours before budgeting.
- Not estimated
- Implementation (6–16 FDE-days)
- 6–16 FDE-days at US$760–1940 per day, converted from the HK$6,000–15,000 band at the HKMA Linked Exchange Rate band of HK$7.75–7.85 to one US dollar. One-off; excludes internal staff time.
- US$4,560 – US$31,040
- An AWS landing zone, an identity provider and a signed agreement are already in place — this adds a service, not a cloud programme.
- The model line is metered and left to the cost section to compute from a fetched price and the usage band; nothing here asserts a per-token figure.
- The knowledge base and vector store recur monthly; implementation is one-off.
05Difficulty
3 / 5
A project week with Linux, Docker and GPU experience
06Skills
- API integrationapi-integration
- development
- AWSaws
- infrastructure
- Compliance and governancecompliance-governance
- compliance
- Identity and SSOidentity-sso
- security
- Networkingnetworking
- networking
- Retrieval-augmented generationrag
- ml
- Security hardeningsecurity-hardening
- security
07Deployment steps
Commands are copied from each project’s own documentation, and the page they came from is linked under the step. 6 of 7 steps currently open an evidence record. The rest are linked to their source; §10 says which of those documents were fetched and which were fetched without their anchor being found — two different states, named differently there.
- 01
Choose the Region, then confirm the model is granted in itversion-sensitive
AssessmentThe Region decides the model catalogue, the processing location and the residency answer. AWS documents that "in each AWS Region where Amazon Bedrock is available, there is one such deployment account per model provider" — so which provider’s model you can call, and where it runs, is a Regional question. Request access to the specific model in the Region your residency rule requires before the design depends on it.
- 02
Set up Bedrock Knowledge Bases for retrieval
AssessmentUse a managed knowledge base for the short path: "when a query is made, a knowledge base searches your data to find relevant information to answer the query", and it can "include citations in the generated response so the original data source can be referenced and accuracy can be checked". Carry your source permissions into the ingestion so retrieval cannot surface a document the asker could not open.
- 03
Or run your own vector store on OpenSearch Serverless
AssessmentThe managed knowledge base holds the vector store for you. Choose the customer-managed path only if you need control over it: AWS describes it as one where "you set up and manage your own RAG pipeline, including the vector store". Amazon OpenSearch Serverless — "an on-demand, auto-scaling configuration for Amazon OpenSearch Service" — is the usual store for that path. It is more to operate, so take it only for a reason.
- 04
Reach Bedrock over a VPC and PrivateLink
AssessmentKeep the model call off the public internet. AWS opens its guidance with the recommendation "to control access to your data, we recommend that you use a virtual private cloud (VPC)", and an interface endpoint (PrivateLink) is what makes the Bedrock call stay on your network. Scope the IAM policy on the model and the knowledge base to least privilege at the same time.
- 05
Build the application, with IAM Identity Center sign-in
AssessmentThe web application, the conversation store and the audit log are yours to build and host. Gate access through IAM Identity Center or your own provider so membership of a group is the only way in, and keep the citations the knowledge base returns next to each answer so a reader can check where it came from.
- 06
Record the DPA scope, the model provider and the routingversion-sensitive
AssessmentWrite down which model you enabled and its provider, that each provider runs in its own per-Region deployment account, the Region every component is pinned to, and the model-invocation logging setting. AWS is a processor; the provider’s terms apply alongside the AWS agreement, and both need to cover the content you will actually send.
- 07
Evaluate on your own documents before launch
AssessmentA benchmark is not evidence about your corpus. Since the knowledge base "searches your data to find relevant information to answer the query", the test is whether it finds the right passage: assemble real questions with agreed answers, run them before launch and after every model or ingestion change, and keep the retrieval hits beside each answer.
08Compliance considerations
Structured issue-spotting to support your own review — not legal advice. Verify against the cited primary sources and your counsel.
Applies everywhere
Data residency · Cross-border transfershigh
Pin the model, the knowledge base, the vector store and the logs to one Region, and check whether any feature routes a request elsewhere. Whether the model you want is offered in your Region at all is the first question for a residency rule.
Data processing agreement · Subprocessors · Model accesshigh
AWS documents that each model provider runs in its own deployment account per Region. Confirm the AWS agreement and the model provider’s terms both cover the use you intend, and that access to the specific model has been granted in your account.
Security · Logginghigh
AWS recommends reaching Bedrock over a VPC. Add PrivateLink so nothing traverses the public internet, scope IAM to least privilege on the model and the knowledge base, and turn on CloudTrail with a retention window that matches your policy.
Retention · Prompt leakagemedium
The retrieved passages are usually the most sensitive part of a prompt. Decide what may be ingested into the knowledge base, keep your own logs of prompts and responses under a stated retention rule, and confirm the model-invocation logging setting is what you intend.
09Alternatives
Azure or Google, if you are on a different cloud
The same design on the cloud you already run: a managed model, a managed retrieval service, your app and identity. Choose by the account you hold, not by the model.
- — Lands in an account, identity provider and agreement you already have
- — Different model catalogue and region availability per cloud
Your own model on GPU inside the same account
When a needed model is not offered, or a residency rule no endpoint satisfies applies, serve open weights on your own EC2 GPU instances rather than calling the managed model.
- — No managed model service sees the prompt; AWS is only an infrastructure processor
- — You own GPU capacity, the serving stack and the evaluation after every upgrade
- Private cloudPrivate-cloud RAG in a single regionA GPU instance in one region running vLLM and Open WebUI, a managed PostgreSQL with pgvector for chats and embeddings, object storage for the original files, and your existing identity provider for sign-in. Same software as the on-premise recipe; different trust boundary.
- Private cloudAssistant on Microsoft Foundry (Azure OpenAI)A model deployed in Microsoft Foundry (the platform that was Azure AI Foundry / Azure OpenAI), Azure AI Search for retrieval over your content, a web application you build, Microsoft Entra ID for sign-in, and private endpoints so the call never leaves your virtual network.
10Evidence
- 01–03Tier 2official vendor legal documentationVendor legal and technical documentation
Retrieve data and generate AI responses with Amazon Bedrock Knowledge Bases
https://docs.aws.amazon.com/bedrock/latest/userguide/knowledge-base.html
FreshRetrieved 27 Aug 2026sha256:1e03b0316b803 records · ·
- 04Tier 2official vendor legal documentationVendor legal and technical documentation
Data protection - Amazon Bedrock
https://docs.aws.amazon.com/bedrock/latest/userguide/data-protection.html
FreshRetrieved 27 Aug 2026sha256:8498ddfbd9581 record
Not yet in the ledger
not yet fetched
not yet fetched
11Community
Deployed this stack, or hit something this page does not cover? Corrections, sources and implementation reports are what keep a recipe worth reading.
Improve this page
Sign in to contribute
From the field
0 deployments · 0 questions
Nobody has reported deploying this here yet, and no question has been opened against this page. Both appear once a reviewer accepts them.
12Hire an FDE
If you would rather not build it, we can introduce a forward-deployed engineer who has deployed this stack before. The enquiry form starts from this recipe.