feat: Add automatic memory provisioning to Bedrock AgentCore CLI #204

sundargthb · 2025-09-24T01:13:50Z

Integrate AgentCore Memory Manager for STM and LTM configuration
Add memory type selection prompt during agent configuration
Implement automatic memory resource creation with appropriate strategies
Support both short-term memory (30-day retention) and long-term memory extraction
Pass memory configuration to container runtime via environment variables
Handle idempotent memory operations for existing resources

Description

This PR integrates AWS Bedrock AgentCore Memory into the CLI toolkit, enabling automatic provisioning of both short-term memory (STM) and long-term memory (LTM) capabilities for agents. Users can now seamlessly enable memory features during agent configuration without manual API calls.

Changes

Configuration Phase: Added memory type selection prompt in configure.py allowing users to choose between STM-only or STM+LTM
Launch Phase: Integrated MemoryManager from operations/memory to automatically provision memory resources
Strategy Management: Automatic creation of three LTM strategies when enabled:
- User Preferences (/users/{actorId}/preferences)
- Semantic Facts (/users/{actorId}/facts)
- Session Summaries (/summaries/{actorId}/{sessionId})
Environment Integration: Memory ID and configuration passed to container runtime

Usage

agentcore configure --entrypoint my_agent.py  # Prompts for memory configuration
agentcore launch                               # Automatically provisions memory
agentcore invoke '{"prompt": "Hello"}'         # Agent has memory capabilities

Type of Change

Bug fix (non-breaking change which fixes an issue)
[X ] New feature (non-breaking change which adds functionality)
Breaking change (fix or feature that would cause existing functionality to not work as expected)
Documentation update
Performance improvement
Code refactoring

Testing

[X ] Unit tests pass locally
[X ] Integration tests pass (if applicable)
[x ] Test coverage remains above 80%
[ x] Manual testing completed

Checklist

[x ] My code follows the project's style guidelines (ruff/pre-commit)
[x ] I have performed a self-review of my own code
[x ] I have commented my code, particularly in hard-to-understand areas
[] I have made corresponding changes to the documentation
[x ] My changes generate no new warnings
[x ] I have added tests that prove my fix is effective or that my feature works
[x ] New and existing unit tests pass locally with my changes
[x ] Any dependent changes have been merged and published

Security Checklist

[x ] No hardcoded secrets or credentials
[x ] No new security warnings from bandit
[x ] Dependencies are from trusted sources
[x ] No sensitive data logged

Breaking Changes

List any breaking changes and migration instructions:

N/A

Additional Notes

Add any additional notes or context about the PR here.

- Integrate AgentCore Memory Manager for STM and LTM configuration - Add memory type selection prompt during agent configuration - Implement automatic memory resource creation with appropriate strategies - Support both short-term memory (30-day retention) and long-term memory extraction - Pass memory configuration to container runtime via environment variables - Handle idempotent memory operations for existing resources

Signed-off-by: Sundar Raghavan <[email protected]>

Remove blocking wait for LTM strategy provisioning during launch Use batch strategy updates to reduce provisioning time from ~3 minutes to seconds Add memory provisioning status to agentcore status command Fix status panel formatting with proper newline characters Use async memory creation for faster agent deployment

… into feature/cli-memory-provisioning

Remove blocking wait for LTM strategy provisioning during launch Use batch strategy updates to reduce provisioning time from ~3 minutes to seconds Add memory provisioning status to agentcore status command Fix status panel formatting with proper newline characters Use async memory creation for faster agent deployment

Signed-off-by: Sundar Raghavan <[email protected]>

bergjaak · 2025-09-26T17:18:53Z

src/bedrock_agentcore_starter_toolkit/utils/runtime/schema.py

+class MemoryConfig(BaseModel):
+    """Memory configuration for BedrockAgentCore."""
+
+    enabled: bool = Field(default=False, description="Whether memory is enabled")


should we enable memory by default? also does this mean short-term memory (event/conversation history) or long-term memory (extraction of memory records from short-term memory)

I see there is also enable_ltm -- I think this is confusing to have both controls. Instead, this argument should be an enum like STM_AND_LTM or STM_ONLY or NO_MEMORY (to have LTM, you must have STM)

short term memory is enabled by default. User will be given a choice to enable LTM - that will be Yes/No - If yes, both STM and LTM - I can clarify the verbiage in setup

Replaced the enabled + enable_ltm boolean combo with a single enum-style field using Literal types (STM_ONLY, STM_AND_LTM, NO_MEMORY)

bergjaak · 2025-09-26T17:34:17Z

src/bedrock_agentcore_starter_toolkit/operations/runtime/configure.py

+
+    # Create memory config
+    memory_config = MemoryConfig()
+    memory_config.enabled = True  # Always true for STM


same comment as below about using enums and having only one input

Replaced the enabled + enable_ltm boolean combo with a single enum-style field using Literal types (STM_ONLY, STM_AND_LTM, NO_MEMORY)

bergjaak · 2025-09-26T17:37:52Z

src/bedrock_agentcore_starter_toolkit/utils/runtime/templates/Dockerfile.j2

+    BEDROCK_AGENTCORE_MEMORY_ID={{ memory_id }}{% endif %}{% if memory_name %} \
+    BEDROCK_AGENTCORE_MEMORY_NAME={{ memory_name }}{% endif %}


do we give the user some easier affordance for setting this up? Like can we (or do we) prompt the user to provide these? Or, better yet, could we call ListMemories and let the user choose and then assist them in creating a memory if there isn't a memory in the list?

Yes - this is planned as a fast follow up

Completed this implementation - Users can choose memory id from the List of memories.

bergjaak · 2025-09-26T17:41:29Z

src/bedrock_agentcore_starter_toolkit/cli/runtime/configuration_manager.py

+        Returns:
+            Tuple of (enable_memory, enable_ltm)
+        """
+        console.print("\\n🧠 [cyan]Memory Configuration[/cyan]")


⚠️ ℹ️ No container engine found (Docker/Finch/Podman not installed) ✅ Default deployment uses CodeBuild (no container engine needed) 💡 Run 'agentcore launch' for cloud-based building and deployment 💡 For local builds, install Docker, Finch, or Podman \n🧠 Memory Configuration ✓ Short-term memory is enabled by default

this \n is extra

thanks, will fix this

still need this :)

bergjaak · 2025-09-26T17:56:05Z

src/bedrock_agentcore_starter_toolkit/operations/runtime/launch.py

+                # Use private method to avoid waiting
+                memory = memory_manager._create_memory(
+                    name=memory_name,
+                    description=f"Memory for agent {agent_name} with {'STM+LTM' if strategies else 'STM only'}",


not sure where to leave this comment, but in agentcore status

➜ agentcore status ✅ MemoryManager initialized for region: us-east-1 ╭────────────────────────────────────────────────────────────────────────── Agent Status: my_agent ──────────────────────────────────────────────────────────────────────────╮ │ Ready - Agent deployed and endpoint available │ │ │ │ Agent Details: │ │ Agent Name: my_agent │ │ Agent ARN: arn:aws:bedrock-agentcore:us-east-1:332634280070:runtime/my_agent-SnemZt47u5 │ │ Endpoint: DEFAULT (READY) │ │ Region: us-east-1 | Account: 332634280070 │ │ │ │ Memory: STM+LTM (3 strategies) (bedrock_agentcore_my_agent_memory-J7Aosv7Rh7) │ │ │ │ Deployment Info: │ │ Created: 2025-09-26 17:50:16.300812+00:00 │ │ Last Updated: 2025-09-26 17:50:17.515972+00:00 │ │ │ │ 📋 CloudWatch Logs: │ │ /aws/bedrock-agentcore/runtimes/my_agent-SnemZt47u5-DEFAULT --log-stream-name-prefix "2025/09/26/[runtime-logs]" │ │ /aws/bedrock-agentcore/runtimes/my_agent-SnemZt47u5-DEFAULT --log-stream-names "otel-rt-logs" │ │ │ │ 🔍 GenAI Observability Dashboard: │ │ https://console.aws.amazon.com/cloudwatch/home?region=us-east-1#gen-ai-observability/agent-core │ │ │ │ ⏱️ Note: Observability data may take up to 10 minutes to appear after first launch │ │ │ │ 💡 Tail logs with: │ │ aws logs tail /aws/bedrock-agentcore/runtimes/my_agent-SnemZt47u5-DEFAULT --log-stream-name-prefix "2025/09/26/[runtime-logs]" --follow │ │ aws logs tail /aws/bedrock-agentcore/runtimes/my_agent-SnemZt47u5-DEFAULT --log-stream-name-prefix "2025/09/26/[runtime-logs]" --since 1h │ │ │ │ Ready to invoke: │ │ agentcore invoke '{"prompt": "Hello"}' │ ╰────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯

we need more details about the memory resource

│ Memory: STM+LTM (3 strategies) (bedrock_agentcore_my_agent_memory-J7Aosv7Rh7)

we should render all the fields that are on the memory resource here. Should be pretty straightforward to just get an LLM to do that imo

…TM memory status on first invoke.

Addresses PR review comments: - Replace boolean flags (enabled/enable_ltm) with single Literal enum type (STM_ONLY, STM_AND_LTM, NO_MEMORY) for clearer memory mode configuration - Add ListMemories integration in configure flow allowing users to select existing memory resources or create new ones with guided prompts - Enhance agentcore status output to display comprehensive memory resource details including all fields, strategies, namespaces, and provisioning status

bergjaak · 2025-09-29T14:41:45Z

src/bedrock_agentcore_starter_toolkit/cli/runtime/configuration_manager.py

+            return False, False
+
+        # If memory is enabled, ask about long-term memory
+        console.print("\\n[dim]Long-term memory extracts:[/dim]")


is the \n a typo here too?

thanks, updated this

bergjaak · 2025-09-29T14:43:05Z

src/bedrock_agentcore_starter_toolkit/cli/runtime/configuration_manager.py

+                return self._prompt_new_memory_config()
+
+            memory_manager = MemoryManager(region_name=region)
+            existing_memories = memory_manager.list_memories(max_results=10)


in a follow up, would be nice to figure out how to list all for pre-selection (like maybe exposed via a search or displayed in 3 columns with the full names truncated)

bergjaak · 2025-09-29T14:43:53Z

src/bedrock_agentcore_starter_toolkit/operations/runtime/configure.py

            else:
                log.debug("No execution role provided and auto-create disabled")

+    # Prompt for memory configuration BEFORE generating Dockerfile


should this commented out code be here?

thanks. updated this

Vivekbhadauria1 · 2025-09-30T00:28:20Z

src/bedrock_agentcore_starter_toolkit/operations/runtime/launch.py

+    # Create memory if configured
+    if agent_config.memory and not agent_config.memory.memory_id:
+        log.info("Creating memory resource for agent: %s", agent_name)
+        try:
+            from ...operations.memory.constants import StrategyType
+            from ...operations.memory.manager import MemoryManager
+
+            memory_manager = MemoryManager(region_name=agent_config.aws.region)
+            memory_name = f"bedrock_agentcore_{agent_name}_memory"
+
+            # Check if memory already exists
+            existing_memory = None
+            try:
+                memories = memory_manager.list_memories()
+                for m in memories:
+                    if m.id.startswith(memory_name):
+                        existing_memory = memory_manager.get_memory(m.id)
+                        log.info("Found existing memory: %s", m.id)
+                        break
+            except Exception as e:
+                log.debug("Error checking for existing memory: %s", e)
+
+            # Determine if we need to create new memory or add strategies to existing
+            if existing_memory:
+                # Check if strategies need to be added
+                existing_strategies = []
+                if hasattr(existing_memory, "strategies") and existing_memory.strategies:
+                    existing_strategies = existing_memory.strategies
+
+                log.info("Existing memory has %d strategies", len(existing_strategies))
+
+                # If LTM is enabled but no strategies exist, add them
+                if agent_config.memory.has_ltm and len(existing_strategies) == 0:
+                    log.info("Adding LTM strategies to existing memory...")
+
+                    # Add all strategies in one batch call
+                    memory_manager.update_memory_strategies_and_wait(
+                        memory_id=existing_memory.id,
+                        add_strategies=[


reuse same logic across code-build and non code-build functions.

Extracted all memory creation logic into a single _ensure_memory_for_agent() helper function at the top of launch.py. Both CodeBuild and non-CodeBuild paths now call this one function.

Vivekbhadauria1 · 2025-09-30T00:33:30Z

src/bedrock_agentcore_starter_toolkit/operations/runtime/launch.py

+            entrypoint_path = Path(entrypoint_parts[0])
+            agent_var_name = entrypoint_parts[1] if len(entrypoint_parts) > 1 else agent_name
+
+            runtime.generate_dockerfile(


We should generate Dockerfile with launch. This should also be dine with non code build launch.

Removed the Dockerfile regeneration in _launch_with_codebuild(). Memory env vars now passed purely at runtime via environmentVariables in the agent deployment call. The _deploy_to_bedrock_agentcore() function already handles this injection

Vivekbhadauria1 · 2025-09-30T00:34:28Z

src/bedrock_agentcore_starter_toolkit/operations/runtime/status.py

        agent_arn=agent_config.bedrock_agentcore.agent_arn,
    )

+    if agent_config.memory and agent_config.memory.memory_id:


memory specific method for status.

Changed status.py to use MemoryManager's dedicated methods instead of raw dictionary access:

get_memory_status(memory_id) - returns status string with proper error handling
get_memory() - returns memory metadata
get_memory_strategies(memory_id) - handles field name variations (strategies vs memoryStrategies)

Vivekbhadauria1 · 2025-09-30T00:36:30Z

documentation/docs/examples/agentcore-quickstart-example.md

+
+## Prerequisites
+
+- **AWS Permissions:** You need specific permissions to use the starter toolkit. See the [Permissions Reference](#permissions-reference) section for the complete IAM policy.


Highlight that if developer is using root user/admins permissions then these steps are not required.

Vivekbhadauria1 · 2025-09-30T00:36:48Z

documentation/docs/examples/agentcore-quickstart-example.md

+        if memory['status'] != 'ACTIVE':
+            agent = Agent(
+                model=MODEL_ID,
+                system_prompt="You are a helpful assistant.",
+                tools=[calculate]
+            )
+            result = agent(payload.get("prompt", ""))
+            return {
+                "response": f"[Memory initializing] {result.message.get('content', [{}])[0].get('text', '')}",
+                "session_id": "temp"
+            }
+    except Exception:
+        pass


remove the check

Vivekbhadauria1 · 2025-09-30T15:24:29Z

documentation/docs/examples/agentcore-quickstart-example.md

+Key components in this implementation:
+
+- **Runtime**: Container orchestration service that hosts your agent
+- **Memory Service**: Dual-layer storage with STM (exact conversation storage) and LTM (intelligent fact extraction)


it does more than fact extraction.

updated to mention all strategies.

Vivekbhadauria1 · 2025-09-30T15:25:02Z

documentation/docs/examples/agentcore-quickstart-example.md

+
+Key components in this implementation:
+
+- **Runtime**: Container orchestration service that hosts your agent


its not an orchestration service. Use descriptions for each aligned aws docs and add those refs.

updated this.

Vivekbhadauria1 · 2025-09-30T15:26:07Z

documentation/docs/examples/agentcore-quickstart-example.md

+    memory_config = AgentCoreMemoryConfig(
+        memory_id=MEMORY_ID,
+        session_id=session_id,
+        actor_id="user",


not use env var for user?

missed to update this in the latest version w/ streaming.. updated this now

Vivekbhadauria1 · 2025-09-30T15:28:19Z

documentation/docs/examples/agentcore-quickstart-example.md

+# "You work at TechCorp."
+```
+
+### Test Code Interpreter with Memory


why is memory important for this?

originally had this testing with different session ids.. removed it

Vivekbhadauria1 · 2025-09-30T15:32:54Z

src/bedrock_agentcore_starter_toolkit/operations/runtime/invoke.py

    project_config = load_config(config_path)
    agent_config = project_config.get_agent_config(agent_name)
+
+    # Check memory status on first invoke if LTM is enabled


why not reuse the status method with status.py instead? seems duplicate logic across invoke and status.

invoke.py needs a single lightweight check: "Is memory ready or should I fail gracefully?" Adding a dependency on status.py would add 2+ API calls per invoke (agent runtime, endpoint, strategies) and couple invoke logic to CLI display formatting

have a method that can serve both?

that will work but needs some design considerations.. right now - invoke.py needs: “Is memory ready?” (boolean, 1 API call). and status.py needs: “Full memory diagnostics” (rich object, 4 API calls).. creating a shared method with a flag like get_memory_info(full_diagnostics=True/False) will returns different types..

Vivekbhadauria1 · 2025-09-30T15:59:39Z

documentation/docs/examples/agentcore-quickstart-example.md

+        code_session_id = code_interpreter.start(
+            name="calc_session",


use same as runtime session id? That would avoid potential crosswiring across multiple sessions/users within an account.

right.. now uses runtime session id

Vivekbhadauria1

Lets follow up on -

Streaming
Memory status simplification
Strands documentation links in sample

Vivekbhadauria1 · 2025-09-30T20:27:15Z

documentation/docs/examples/agentcore-quickstart-example.md

+- **Runtime**: Managed compute service that runs your containerized agent with automatic scaling. See [AgentCore Runtime docs](https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/agents-tools-runtime.html)
+- **Memory Service**: Dual-layer storage with short-term memory (chronological event storage with 30-day retention) and long-term memory (extraction of user preferences, semantic facts, and session summaries). See [AgentCore Memory docs](https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/memory.html)
+- **Code Interpreter**: AWS-managed Python sandbox with pre-installed libraries. See [AgentCore Code Interpreter](https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/code-interpreter-tool.html)
+- **Strands Framework**: Simplifies agent creation with memory session management


ref for this and the description is not correct.

sundargthb requested a review from Vivekbhadauria1 September 24, 2025 01:24

Sundar Raghavan and others added 2 commits September 23, 2025 18:32

fix: resolve conflict

5f3b3be

Merge branch 'main' into feature/cli-memory-provisioning

1331eab

Signed-off-by: Sundar Raghavan <[email protected]>

sundargthb had a problem deploying to auto-approve September 24, 2025 01:36 — with GitHub Actions Error

Sundar Raghavan added 3 commits September 25, 2025 06:16

Merge remote-tracking branch 'origin/feature/cli-memory-provisioning'…

615099e

… into feature/cli-memory-provisioning

sundargthb had a problem deploying to auto-approve September 25, 2025 13:51 — with GitHub Actions Error

complete unit test files

2235370

sundargthb had a problem deploying to auto-approve September 25, 2025 15:33 — with GitHub Actions Error

Sundar Raghavan and others added 5 commits September 25, 2025 11:05

improve test coverage

973cefc

improve test coverage

ecb6078

resolve conflicts

d6c49c0

resolve conflicts

6580075

Merge branch 'main' into feature/cli-memory-provisioning

36003c1

Signed-off-by: Sundar Raghavan <[email protected]>

sundargthb had a problem deploying to auto-approve September 25, 2025 22:16 — with GitHub Actions Error

sundargthb requested review from jona62, bergjaak and siwachabhi September 26, 2025 17:13

bergjaak reviewed Sep 26, 2025

View reviewed changes

docs: add quickstart example

15d62fc

sundargthb had a problem deploying to auto-approve September 26, 2025 22:57 — with GitHub Actions Error

update execution role policy for code interpreter

af9987d

sundargthb had a problem deploying to auto-approve September 27, 2025 16:25 — with GitHub Actions Error

feat: Add graceful memory provisioning check on first invoke. Check L…

ec286bf

…TM memory status on first invoke.

sundargthb had a problem deploying to auto-approve September 27, 2025 21:08 — with GitHub Actions Error

sundargthb had a problem deploying to auto-approve September 28, 2025 20:06 — with GitHub Actions Error

update unit test files

0c77586

sundargthb had a problem deploying to auto-approve September 28, 2025 23:31 — with GitHub Actions Error

bergjaak reviewed Sep 29, 2025

View reviewed changes

fix formatting errors

32cad80

sundargthb had a problem deploying to auto-approve September 29, 2025 14:59 — with GitHub Actions Error

update quickstart example

398c0db

sundargthb had a problem deploying to auto-approve September 29, 2025 16:51 — with GitHub Actions Error

add non-interactive for memory

9b0d484

sundargthb had a problem deploying to auto-approve September 29, 2025 17:31 — with GitHub Actions Error

fix lint error

446647f

sundargthb had a problem deploying to auto-approve September 29, 2025 22:46 — with GitHub Actions Error

Vivekbhadauria1 requested changes Sep 30, 2025

View reviewed changes

complete PR feedback updates

7f7f6bb

sundargthb had a problem deploying to auto-approve September 30, 2025 02:59 — with GitHub Actions Error

sundargthb requested review from Vivekbhadauria1 and removed request for siwachabhi September 30, 2025 13:48

Vivekbhadauria1 requested changes Sep 30, 2025

View reviewed changes

complete PR feedback updates

19875db

sundargthb had a problem deploying to auto-approve September 30, 2025 20:23 — with GitHub Actions Error

Vivekbhadauria1 approved these changes Sep 30, 2025

View reviewed changes

sundargthb merged commit d58b61c into main Sep 30, 2025
16 of 17 checks passed

		BEDROCK_AGENTCORE_MEMORY_ID={{ memory_id }}{% endif %}{% if memory_name %} \
		BEDROCK_AGENTCORE_MEMORY_NAME={{ memory_name }}{% endif %}


		## Prerequisites

		- AWS Permissions: You need specific permissions to use the starter toolkit. See the [Permissions Reference](#permissions-reference) section for the complete IAM policy.


		Key components in this implementation:

		- Runtime: Container orchestration service that hosts your agent

		code_session_id = code_interpreter.start(
		name="calc_session",

feat: Add automatic memory provisioning to Bedrock AgentCore CLI #204

feat: Add automatic memory provisioning to Bedrock AgentCore CLI #204

Uh oh!

Conversation

sundargthb commented Sep 24, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

Changes

Usage

Type of Change

Testing

Checklist

Security Checklist

Breaking Changes

Additional Notes

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

sundargthb commented Sep 24, 2025 •

edited

Loading