Migrate

Input

$ARGUMENTS is optional:

What migration does

The agentcore create --type import command reads your existing Bedrock Agent’s configuration and generates an AgentCore project that reproduces its behavior in a code-first framework. Specifically:

• System prompt → copied into the generated main.py
• Action groups (Lambda-backed) → converted to Gateway targets with --type lambda-function-arn
• Knowledge bases → referenced in the system prompt with a note to wire retrieval manually (AgentCore doesn’t auto-import KB bindings)
• Guardrails → noted in comments but not auto-converted (AgentCore uses Cedar policies, not Bedrock Guardrails)
• Agent alias / version → the import targets a specific alias, not the draft

What migration does not do:

• It does not delete or modify the original Bedrock Agent — the source agent keeps running
• It does not migrate conversation history or session state
• It does not convert Bedrock Guardrails to Cedar policies (different authorization model)
• It does not auto-wire Knowledge Base retrieval — you’ll need to add that as a tool or direct SDK call

Prerequisites

1. The Bedrock Agent must exist and have at least one alias
2. Your AWS credentials must have bedrock:GetAgent, bedrock:GetAgentAlias, and bedrock:ListAgentActionGroups permissions
3. You need the agent ID, alias ID, and region

Process

Step 1: Run the import

The --framework flag determines which code-first framework the generated project uses. Strands is recommended for the closest mapping to Bedrock Agent behavior.

Project name rules apply: max 23 characters, alphanumeric only, starts with a letter.

Step 2: Review the generated project

Check:

• The system prompt matches your original agent’s instructions
• Action groups appear as Gateway targets in agentcore.json (under agentCoreGateways)
• The model ID is correct for your target region

Step 3: Fill in what migration doesn’t cover

Knowledge Bases: If your Bedrock Agent used Knowledge Bases, you have two options:

1. Keep using the KB via boto3 — call bedrock-agent-runtime:RetrieveAndGenerate or Retrieve directly from your agent code as a tool
2. Replace with AgentCore Memory — if the KB was used for user-specific context, AgentCore Memory with SEMANTIC strategy may be a better fit. See memory.md.

Guardrails → Cedar policies: Bedrock Guardrails (content filters, denied topics, word filters) don’t have a 1:1 mapping to Cedar policies. Cedar policies control which tools the agent can call and with what parameters — they’re authorization rules, not content filters. If you need content filtering, keep the guardrail logic in your agent code (pre/post-processing) or use Bedrock Guardrails as a standalone API call.

Custom orchestration: If your Bedrock Agent used custom orchestration (return-of-control, custom Lambda orchestrators), you’ll need to rebuild that logic in the framework’s native patterns — Strands tool chains, LangGraph graph nodes, etc.

Step 4: Test locally and deploy

Step 5: Cut over traffic

Once the AgentCore agent is working correctly:

1. Update your application to invoke the AgentCore Runtime instead of the Bedrock Agent
2. See integrate.md for the invocation patterns (SigV4, JWT, SDK)
3. Keep the original Bedrock Agent running as a fallback until you’re confident
4. Delete the Bedrock Agent only after the AgentCore agent has been stable in production

Common migration issues

“Model not available in target region” The imported agent may reference a model ID that isn’t available in your AgentCore deployment region. Edit model/load.py to use a cross-region inference profile or a model available in your region.

“Action group Lambda in a different region” Gateway targets can invoke Lambda functions cross-region, but latency increases. Consider deploying the Lambda in the same region as your AgentCore agent, or accept the latency trade-off.

“Agent behavior differs after migration” The most common cause is prompt format differences between Bedrock Agent’s orchestration and the code-first framework. Bedrock Agent injects structured XML around tool results; Strands/LangGraph use different formats. Tune the system prompt to compensate.

Output

• A working AgentCore project that reproduces the Bedrock Agent’s behavior
• A list of what was auto-converted and what needs manual work
• Guidance on cutting over traffic from the old agent to the new one