AWS AgentCore

Enterprise Feature

AWS AgentCore is an enterprise feature. Contact EPAM to verify it is enabled for the deployment.

AWS AgentCore integration enables the discovery of AWS-hosted agent runtimes, browsing their endpoints, and installing those endpoints as assistants directly within the platform — without writing any integration code.

Prerequisites

An active AWS integration must be configured before any AgentCore runtimes are visible. See the AWS Integration guide.

Navigating to AgentCore Runtimes

To access AgentCore management, open Settings from the left navigation bar and navigate to External Vendors → AWS Integration.

Once an AWS integration is configured, the Manage AgentCore Runtimes table reflects its discovery status:

Column	Description
Project Name	The workspace user/project the integration belongs to.
Runtimes	Discovery result — shows a runtime name (with the AWS logo) when found, No runtimes found when the integration is healthy but returns nothing, or Connection Error (red) when credentials or region are misconfigured.
Integration	The integration identifier and timestamp, or a friendly alias such as aws_env or amazon.
Actions	Manage — opens the integration configuration to update credentials or remove it. The ⓘ icon shows integration metadata.

Connection Error

A Connection Error status means the platform cannot reach the AWS endpoint with the stored credentials. Re-open Manage and verify the access key, secret, and region.

Click AgentCore Runtimes to open the runtime list for the active project.

Browsing Runtimes

The AgentCore Runtimes page shows one card per runtime discovered from the connected AWS account.

Each card displays:

Field	Meaning
Name	Human-readable runtime name as defined in AWS.
ID	Stable unique identifier for the runtime.
AgentCore Runtime	The underlying AgentCore resource name in AWS.
Status badge	Lifecycle state — READY means the runtime is accepting traffic.
Version badge	The runtime schema/protocol version.

Click View on any card to open the Runtime Details page.

Viewing Runtime Details

The Runtime Details page provides a full view of a single runtime and all its endpoints.

Header

Field	Description
Name	Runtime name as defined in AWS.
ID	Stable unique runtime identifier.
Last updated	Date the runtime record was last refreshed.
Status	Lifecycle state of the runtime.
Version	Protocol version.
Source label	AgentCore Runtime: <runtime-name>

Endpoints

Each runtime exposes one or more named endpoints. For each endpoint the table shows:

Column	Meaning
Name	Endpoint label as defined in AWS (can be environment-specific).
ID	Stable endpoint identifier.
Status	Whether this specific endpoint is live.
Version	Protocol version of the endpoint.

Endpoint Actions

Button	Description
Uninstall	Removes the assistant previously installed from this endpoint. Visible only when the endpoint is already installed.
Configure	Opens the Configure endpoint dialog to install or update the assistant mapping.
Chat	Opens a quick chat session directly against this endpoint for testing.
ⓘ	Shows raw endpoint metadata (IDs, ARNs, region info).

Installing an Endpoint as an Assistant

Clicking Configure on any endpoint opens the Configure endpoint: <ENDPOINT_NAME> dialog. Saving this dialog installs — or reconfigures — the endpoint as a named assistant available across the workspace.

Assistant

Field	Required	Default	Description
Assistant Name	No	<runtimeId>:<endpointId>	Display name in the assistant picker.
Assistant Description	No	Auto-generated	Description shown in the assistant list.

Request

Message Path

Field	Required	Default	Description
Message Path	No	message	Dot-notation path where the user message is placed in the request body. For example, message produces { "message": "Hello" }; input.text produces { "input": { "text": "Hello" } }.

History (expandable)

Omit this section entirely to never send conversation history to the endpoint.

Field	Required	Default	Description
History Path	Yes	—	Dot-notation path where the turns array is injected (e.g. messages).
Role Path	No	role	Field name for the role inside each turn object.
Message Path	No	content	Field name for the message text inside each turn object.
User Role	No	user	Role label used for user turns.
Assistant Role	No	assistant	Role label used for all non-user turns.

Extra Payload (expandable)

A static JSON object whose keys are merged into every request body. Useful for runtime-specific parameters such as session_id, locale, or model overrides. Must be a flat or nested object — arrays and scalar values are rejected.

Response

Field	Required	Default	Description
Enable Streaming	No	Off	When enabled, the platform reads the response as an SSE stream and displays tokens incrementally. When disabled, the platform reads a single JSON body.
Response Text Path	Yes (when streaming off)	—	Dot-notation path to the answer text in the full response body (e.g. output).
Chunk Text Path	Yes (when streaming on)	—	Dot-notation path to the answer text in each SSE chunk (e.g. delta.content).

Thought Extraction (expandable)

If the AgentCore runtime surfaces chain-of-thought reasoning separately from the final answer, configure extraction here.

Field	Required	Description
Text Path	Yes	Dot-notation path to the thought text.
Thoughts Path	No (non-streaming)	Path to an array of thought objects; Text Path, Name Path, and Args Path are resolved per item in the array.
Name Path	No	Path to the tool or agent name within each thought object.
Args Path	No	Path to the tool arguments within each thought object.

Saving

Click Save to apply. The endpoint status on the Runtime Details page reflects the installed assistant immediately. To update an already-installed endpoint, click Configure again — the dialog reopens pre-populated with the current values. Click Cancel to discard changes.

AgentCore Agent Implementation Examples

The examples below show the exact request and response shapes the AgentCore agent must implement for each configuration.

Streaming — with history and thought extraction

Configure endpoint settings

Field	Value
Message Path	message
History Path	messages
Role Path	role
History Message Path	content
User Role	user
Assistant Role	assistant
Enable Streaming	On
Chunk Text Path	delta.text
Thought Extraction — Text Path	delta.thinking
Thought Extraction — Active Path	delta.thinking_active

Request body sent to AgentCore

{
  "message": "What is the capital of France?",
  "messages": [
    { "role": "user", "content": "Hi there" },
    { "role": "assistant", "content": "Hello! How can I help?" }
  ]
}

SSE chunks the agent must return

{ "delta": { "text": null, "thinking": "User is asking geography.", "thinking_active": true } }
{ "delta": { "text": null, "thinking": null, "thinking_active": false } }
{ "delta": { "text": "The capital of France is Paris.", "thinking": null, "thinking_active": false } }

---

Non-streaming — with history and thought extraction

Configure endpoint settings

Field	Value
Message Path	input.text
History Path	input.history
Role Path	role
History Message Path	content
User Role	human
Assistant Role	bot
Enable Streaming	Off
Response Text Path	output.answer
Thought Extraction — Thoughts Path	output.thoughts
Thought Extraction — Text Path	content
Thought Extraction — Name Path	tool
Thought Extraction — Args Path	input

Request body sent to AgentCore

{
  "input": {
    "text": "What is the capital of France?",
    "history": [
      { "role": "human", "content": "Hi there" },
      { "role": "bot", "content": "Hello! How can I help?" }
    ]
  }
}

Response body the agent must return

{
  "output": {
    "answer": "The capital of France is Paris.",
    "thoughts": [
      { "tool": "knowledge_base", "input": { "query": "capital France" }, "content": "Retrieved: Paris is the capital." },
      { "tool": null, "input": null, "content": "Sufficient confidence, no further lookup needed." }
    ]
  }
}