Key takeaways
- AgentCore provides Runtime, Gateway, Identity, Memory, Observability, a Browser Tool and a Code Interpreter, which can be used on their own or together, with any open-source framework and foundation model.
- Pricing is consumption-based with no upfront commitments or minimum fees. Runtime is billed per second for the CPU and memory actually used, and CPU scales to zero while an agent waits on models or tools.
- AgentCore suits agents you build yourself and run in your own AWS account; vendor agents such as Copilot Studio or Agentforce live inside those vendors' applications.
What AgentCore is
AWS made Amazon Bedrock AgentCore generally available in October 2025. It is a set of services that handle what an agent needs in production, from a secure runtime to identity, memory, tool access and monitoring, so teams don't have to build that infrastructure themselves. AWS says it works with any open-source framework and any foundation model.
The services
| Service | What it does |
|---|---|
| Runtime | A secure, serverless environment for deploying and scaling agents and tools, with session isolation. |
| Gateway | Turns APIs and Lambda functions into agent tools and connects to existing MCP servers, as one secure endpoint for tools. |
| Identity | Lets agents act on behalf of users or by themselves, with secure storage of tokens and access control. |
| Memory | Session and long-term memory, so agents keep context. |
| Observability | Visibility into agent performance and behaviour for debugging, audit and monitoring. |
| Browser Tool and Code Interpreter | A cloud browser for agents to use websites, and a sandbox to run code. |
| Harness | A production-ready agent behind two API calls, using the services above; no charge for the harness itself. |
How AgentCore is priced
- Consumption-based, with no upfront commitments or minimum fees; you pay for each service you use.
- Runtime: billed per second on actual CPU and memory consumption, with CPU scaling to zero during I/O wait; a committed baseline gives a discounted rate.
- Identity: no additional charge when used through AgentCore Runtime or Gateway.
- Model inference: billed separately at your model provider's rates.
How to start
Define the agent
What it does, for whom, which systems it uses and what it must never do.
Expose the tools
Put your APIs and functions behind AgentCore Gateway, with identity for each agent.
Build and test
Build with your chosen framework and model, and test with real tasks and Observability on.
Estimate the cost
Measure runtime and model use in a pilot, then size the production budget.
Launch and monitor
Go live for one group of users and review traces, errors and cost each month.
What to check before you start
- Region: confirm the AWS Regions where the AgentCore services you need are available, against your data rules and the law that applies to you.
- Model: choose a model that fits the language, quality and cost you need, including Arabic if your users need it.
- Least privilege: give each agent only the tools and access its job needs.
- Cost controls: budgets and alerts for runtime and model use.
How Birbal AI helps
Birbal AI builds and runs agents on Amazon Bedrock AgentCore with engineers based in the UAE, and we also provide AI developers who work onsite or remotely with your team. Birbal AI is an independent company, not part of AWS.
Questions and answers
What is Amazon Bedrock AgentCore?
A set of AWS services for deploying and operating AI agents securely at scale: Runtime, Gateway, Identity, Memory, Observability, a Browser Tool and a Code Interpreter, with any framework or model.
How much does AgentCore cost?
AWS prices it by consumption with no upfront commitments or minimum fees. Runtime is billed per second for the CPU and memory actually used; model inference is billed separately.
Does AgentCore only work with Amazon's models?
No. AWS says it works with any open-source framework and foundation model.
What is AgentCore Gateway?
A service that turns APIs and Lambda functions into tools agents can use and connects to existing MCP servers, giving agents one secure endpoint for tools.
When should we use AgentCore instead of vendor agents?
When you are building your own agent, for a product or a process that spans many systems, and want it to run in your own AWS account. For work inside one vendor's application, that vendor's agents are often the quicker start.
Can Birbal AI build AgentCore agents for us?
Yes, with engineers based in the UAE. Ask for a free estimate and you get a scope, timeline and cost range within 48 hours.
Sources
- Amazon Bedrock AgentCore FAQs (AWS)
- Amazon Bedrock AgentCore pricing (AWS)
- Amazon Bedrock AgentCore is now generally available (AWS)
- AgentCore harness: observability and cost controls (AWS documentation)
- Amazon Bedrock AgentCore (AWS Prescriptive Guidance)
Birbal AI is an independent company based in Abu Dhabi. Facts about third parties come from the sources above; check them for the latest position.