Amazon Bedrock models
One API for streaming and tools across models, kept inside one geography such as the EU.
AWS Bedrock agents development
AI agents on AWS.
In your account, under your controls.
Claude, Llama, Mistral and Nova inside your AWS account, with your permissions and regions. We design the agent, its data, and its voice and phone channels.
Ultrabot, built by MLAIA Data ScienceLed by Dr. Yochai Edlitz · Ph.D., Weizmann Institute
The building blocks
Pick the managed pieces.
Write code where the logic is yours.
Managed agent features or your own code: it depends on how much control you need.
Bedrock Agents & AgentCore
Managed agents, or AgentCore to run your own agent code with managed memory and identity.
Knowledge Bases & retrieval
Managed search over S3 documents, or OpenSearch or pgvector when you need custom permissions.
Guardrails
Content filters, denied topics and masking of sensitive data, alongside tool permissions.
Lambda & Step Functions
Tools as Lambdas with narrow roles; long processes with retries and human approval.
Your own agent code
Full control over each step, still in your account. See LangGraph agents.
Voice & contact centre on AWS
The same agent,
speaking on the web or the phone.
Transcribe (including Hebrew), Polly, Nova Sonic and Connect cover voice end to end. In a healthcare assistant we are developing, everything runs EU-only and the code refuses any other region. More on voice agents.
Web voice
Browser audio in, a voice streamed back sentence by sentence.
Phone
Amazon Connect, with transfer to a person and a summary.
Speech-to-speech
Nova Sonic in one step, compared with a pipeline first.
Services we work withBedrockAgentCoreKnowledge BasesGuardrailsLambdaStep FunctionsECS FargateAPI Gateway WebSocketTranscribePollyNova SonicConnectS3OpenSearchSecrets ManagerCloudWatch
Security & cost
Least privilege for tools.
Hard limits for spend.
The risk is in what tools can touch and what the agent can spend. We set both.
Narrow roles per tool
Each tool gets its own narrow IAM role. No broad credentials.
Region & data residency
Pinned to your allowed regions; prompts and audio kept only if you choose.
Budgets & throttles
Rate limits, caps per turn, and budget alarms that can switch off a public endpoint.
Audit trail
Every tool call logged, without storing the user's words.
A practical delivery path
Start small in your account.
Then harden and scale.
Choose the use case
Users, channels, actions, data and regions.
Prototype on Bedrock
Compare models on your tasks; managed or own code.
Tools & channels
Narrow-role tools, retrieval, voice and approvals.
Harden & operate
Infrastructure as code, alarms and tests on every change.
Before we start
Good questions. Straight answers.
Does our data leave our AWS account when using Bedrock?
Bedrock runs models within AWS, and you choose the regions. Data handling depends on the model provider and your configuration; we document which services, regions and logs each request touches before go-live.
Bedrock Agents, AgentCore or our own code?
Managed agents are quickest for standard tool-calling assistants. AgentCore suits teams that want their own agent code with a managed runtime. Fully custom code on Lambda or containers gives the most control. We often prototype with one and keep the option to move.
Can the agent answer phone calls?
Yes. Amazon Connect can route calls to the agent through streaming speech services, with transfer to a person when needed. We measure end-to-end delay on real phone audio, which is narrower-band than web audio.
Does it support Hebrew?
Amazon Transcribe supports Hebrew streaming transcription and Bedrock models handle Hebrew text. Speech output in Hebrew usually needs a voice from another provider, which we connect alongside AWS.
How do you control cost for a public-facing agent?
Rate limits per user and session, short sessions, caps on tokens and tool calls per turn, a bot check before expensive steps, and a budget alarm that turns the endpoint off at a fixed monthly amount.
Talk to a person
Tell us what
you’re working on.
Share the problem, the data you have and what success would look like. We’ll discuss a practical next step.
Please don’t include confidential datasets, credentials or patient information.