Cloud giant Amazon Web Services (AWS) He believes that AI agents will change the way all of us work and interact with information, and enterprises need a platform that enables them to build and implement agents on a large scale – every thing in one place.
Today at the top in New York, AWS presented Amazon Bedrock AgentcoreIN A new corporate class platform designed for building, implementing and servicing AI agents safely and on a scale.
Swami Sivasubraman, Vice President of AWS for Agentic AI, said during the fundamental violence that Agentcore “helps organizations go beyond experiments to agent systems ready for production, which can be trusted in the most critical business processes.”
Agentcore is a modular pile of services – available in preview – which provides programmers the basic infrastructure needed to transfer AI agents from prototype for production, including the executive environment, memory, identity, remark, API integration and tools for browsing web sites and code.
The AI Impact series returns to San Francisco – August 5
The next AI phase is here – are you ready? Join the leaders from Block, GSK and SAP to see the exclusive look at how autonomous agents transform the flows of the work of the company-decision-making in real time for comprehensive automation.
Secure your home now – the space is limited: https://bit.ly/3guplf
“We think that agents will generally change the way we use tools and the Internet,” said Deepak Singh, AWS Vice President of AWS database and AI. “The line between the agent and the application becomes blurred.”
Agentcore is based on the existing framework of Bedrock agents, launched at the end of 2024, but radically expands the possibilities, serving any agent’s funds or a foundation model – not only those who are hosted in Bedrock.
It covers Compatibility with open source tools, corresponding to Cretai, Langchain, Llamaandex, Langraph and Own Strands Agents SDK.
What AWS Bedrock Agentcore includes
- Agentcore Runtime: Without a low delay server, which supports multimodal loads and long -lasting session sessions.
- Agentcore memory: Long -term and short -term memory services that allow agents to learn from past interaction and maintain contextual knowledge in various sessions.
- Agent’s identity: Identity and access management based on Oauth, enabling agents to act on behalf of users in systems corresponding to GitHub, Slack or Salesforce.
- Observable Agentcore: Built -in navigation desktops, debugging and telemetry tools with opentemetry, Langsmith and Datadog support.
- Agentcore Gateway: It transforms internal API interfaces, lambda functions and third -party services into tools compatible with agents using the contextual protocol of the model (MCP).
- Agentcore browser: It provides access to a browser for agents for autonomous interaction with web sites.
- Agentcore code interpreter: Safe environment for performing code generated by agents for evaluation and visualization.
Agentcore also integrates with the AWS market, enabling teams to find and implement pre -built agents and tools.
According to Singh, Agentcore was designed for interoperability. Supports new industry standards, corresponding to the MCP and Google’s Agent-2-Agent (A2A) protocol. Functions corresponding to the AGENTCORE identity and gate are provided by agents by a clear permit and can safely affect internal systems and third parties API interfaces.
The premiere of AWS places it directly in the center of what quickly becomes one of the best segments in Enterprise AI.
Openai SDK agents i Agents based on Gemini Google SDK Both push similar visions of comprehensive programming platforms of agents.
AI networks and a author’s startups corresponding to Cognition (Maker of Devin) are also tools for managing autonomous software agents.
“Agents are the most influential change we have seen for centuries,” said Sivasubramanian. “Together with agents, he comes to services as software. It is a tectonic change in the method of building, implementing and operating software.”
Receiving customers and early cases of use
Several companies have awarded early access to Agentcore, they are already building Agentic Applications in all industries, including finance, healthcare, marketing and content management.
Documents in the cloud and boxes with files storage memory examines expand content management tools using Strands agents and Bedrock Agentcore.
Cto Ben Kus said that integration gives clients boxing “security and compliance with the highest level” when scaling AI’s capabilities in corporate environments.
ITAú Unibanco from Brazil uses Agentcore to support the development of hyper-personal, protected digital banking. The Carlos Eduardo Mazza technology director said that the new platform “will help us provide intuitive banking experience in the scope of automation efficiency and expected personalization.”
In the healthcare space Innovaccer built a new protocol – HMCP (contextual protocol of the healthcare model) – at the summit of Agentcore Gateway. The CEO and co -founder of ABHINAV Shashank called Gateway a “changing game”, which allows the company to remodel existing API interfaces into tools compatible with large -scale agents while maintaining trust, compliance and operational performance.
The Epsilon marketing company uses Agentcore to hurry up the campaign building time and improve your commitment. Prashanth Athota, SVP of Software Engineering, said that the company expected to shorten the construction time by as much as 30% and improve the personalization of customer travel.
Availability and prices
Agentcore is now available in viewing in Select AWS Regions, including US East (N. Virginia), US West (Oregon), Asia Pacific (Sydney) and Europe (Frankfurt). Until September 16, 2025, it is free and prices will start later.
Prices for Agentcore are completely based on consumption, without obligations in advance or minimal fees. Each module – rental, memory, identity, remark, gate, browser and code translator – is settled independently and could be used a la carte or together.
Execution services, browsers and code translator are valued at a second, based on processor and memory consumption, with rates set $ 0.0895 per VCPU-hour and 0.00945 USD for GB.
Fees for the gate 0.005 USD for 1000 API of tools, 0.025 USD for 1000 search inquiries and USD 0.02 for 100 indexed tools per 30 days.
Memorial costs are based on the size of the data: USD 0.25 for 1000 short -term memory events, 0.75 USD for 1000 long -term stored memories (or 0.25 USD with custom strategies) and 0.50 USD for 1000 download.
Agentcore identity costs $ 0.010 for 1000 tokens or API requests, although it is concluded without additional fees when it is used for executive time or gate.
Observable is settled through the Amazon Cloudwatch rates.
To learn more or start, AWS directs programmers to him Agentcore documentation, GitHub samples and devoted Server of disagreement.