What can an AI agent do with AWS Knowledge Bases?

An agent can query your Bedrock Knowledge Bases with natural language, retrieve semantically ranked document chunks, list all configured data sources, trigger ingestion syncs to keep your index fresh, and monitor job status, all automatically on a schedule or on demand. Write actions like syncing require your approval.

Do I need to manage embeddings or vector stores myself?

No. Amazon Bedrock Knowledge Bases handles chunking, embedding, and vector storage for you. Your AI agent interacts with the managed API, so you get full RAG capabilities without managing infrastructure.

Which data sources does this support?

Any data source you've connected to a Bedrock Knowledge Base: including S3 buckets, Confluence, SharePoint, Salesforce, and web crawlers. Your agent queries whatever is indexed in Bedrock.

Is retrieved document content stored?

Document content is accessed during task execution but not permanently stored. Only task outputs such as generated answers, summaries, or reports are saved to your run history.

Developer tools

AWS Knowledge Bases, run by an agent.

An AI agent that queries and manages your Amazon Bedrock Knowledge Bases to retrieve precise answers from your own data.

Connect AWS Knowledge Bases Book a demo

What needs your approval

AWS Knowledge Bases never changes anything without you.

Some read-only AWS Knowledge Bases operations run on their own so your agent can gather what it needs. Anything that creates, edits or deletes waits for your approval.

A real run

An example of how a run works, including the approval step. Not a AWS Knowledge Bases run.

Google Ads

Paid search

Working

review last week's google ads and cut wasted spend.

Queried Google Ads · last 7 days · 4 campaigns

Analyzed search terms · 312 terms · 18 with zero conversions

Drafted changes · 12 negative keywords · 2 budget shifts

Found $1,840/mo going to search terms that never convert. I'd add 12 negatives and move that budget to your two best campaigns. This changes the account, so it's your call before I touch anything.

Plan approval

Add 12 negative keywords stops $1,840/mo on terms with 0 conversions · reversible
Shift budget move the freed spend to your 2 best campaigns · roas 4.2×

Waiting for your approval. Nothing runs until a human decides.

example

Connecting AWS Knowledge Bases

About 5 minutes

Auth: API key
Setup: About 5 minutes

Where to get access

AWS Docs MCP is a free, public service for searching AWS documentation. No API key required — enter any value (e.g. 'enabled') to activate.

github.com

Example tasks

Real work people hand to an agent that uses AWS Knowledge Bases.

run rag queries over internal documentation and return cited answers in Slack
schedule nightly syncs of S3 data sources and alert on ingestion failures
build an enterprise search assistant that queries multiple knowledge bases and ranks results
automate document Q&A pipelines that retrieve context and generate answers with full source attribution