Deploy AI agents to production with a single git push — this project demonstrates true continuous deployment for AWS Bedrock AgentCore agents using GitHub Actions.

Every commit to main automatically:

• 🚀 Deploys a new version of each agent to AWS Bedrock
• 🏗️ Builds and pushes Docker images to Amazon ECR
• 📦 Tracks deployment history via timestamped artifacts
• ⚡ Runs in parallel for multi-agent deployments

No manual intervention required. Push your code, and your agents are live in minutes.

Each push to main triggers a full deployment cycle:

git push origin main
      ↓
GitHub Actions detects the push
      ↓
Workflow extracts all agents from .bedrock_agentcore.yaml
      ↓
Parallel deployment begins for each agent
      ↓
Docker image built → pushed to ECR → deployed to Bedrock
      ↓
New version is live! ✨

The deployment triggers on:

• Direct pushes to the main branch
• Merged pull requests to main

The GitHub Actions workflow (.github/workflows/deploy-agent.yml) uses a matrix strategy to deploy multiple agents efficiently:

Setup Job:

1. Extracts all agent names from .bedrock_agentcore.yaml using yq
2. Installs Python and uv package manager
3. Installs dependencies once (uv sync)
4. Caches the environment for reuse

Deploy Jobs (parallel):

1. Restores the cached environment
2. Configures AWS credentials
3. Runs agentcore launch --agent {agent-name} for each agent
4. Saves agent-specific .bedrock_agentcore.yaml as an artifact

This approach eliminates redundant dependency installation, significantly speeding up multi-agent deployments.

You need to configure AWS authentication for GitHub Actions. Two options:

Option 1: OIDC (Recommended)

1. Create an IAM OIDC identity provider for GitHub Actions:

   Provider URL: https://token.actions.githubusercontent.com
   Audience: sts.amazonaws.com
   

2. Create an IAM role with trust policy:

   {
     "Version": "2012-10-17",
     "Statement": [
       {
         "Effect": "Allow",
         "Principal": {
           "Federated": "arn:aws:iam::YOUR_ACCOUNT:oidc-provider/token.actions.githubusercontent.com"
         },
         "Action": "sts:AssumeRoleWithWebIdentity",
         "Condition": {
           "StringEquals": {
             "token.actions.githubusercontent.com:aud": "sts.amazonaws.com",
             "token.actions.githubusercontent.com:sub": "repo:YOUR_ORG/YOUR_REPO:ref:refs/heads/main"
           }
         }
       }
     ]
   }

3. Attach necessary permissions to the role:

   • Amazon ECR (push images)
   • AWS CodeBuild (trigger builds)
   • Amazon Bedrock AgentCore (deploy agent)

4. Add the role ARN as a GitHub secret: AWS_ROLE_ARN

Option 2: Access Keys

1. Create an IAM user with programmatic access
2. Attach necessary permissions (same as above)
3. Add GitHub secrets:
   • AWS_ACCESS_KEY_ID
   • AWS_SECRET_ACCESS_KEY

To deploy a specific agent:

uv sync
uv run agentcore launch --agent <agent-name>

• Python 3.12+
• uv package manager
• AWS credentials configured

# Install uv
curl -LsSf https://astral.sh/uv/install.sh | sh

# Install dependencies
uv sync

# Run the agent locally (if applicable)
uv run python agent.py

You can invoke the agent locally for testing:

curl -X POST http://localhost:8080/invocations \
  -H "Content-Type: application/json" \
  -d '{"prompt": "Hello!"}'

Adding a new agent is simple and automatically deployed:

# Configure a new agent interactively
uv run agentcore configure --agent <new-agent-name>

# Commit and push
git add .bedrock_agentcore.yaml
git commit -m "Add new agent"
git push origin main

What happens next:

1. Configure command updates .bedrock_agentcore.yaml with your new agent
2. Push to main triggers the deployment workflow
3. Workflow automatically detects the new agent
4. New agent is built and deployed alongside existing agents in parallel
5. Your new agent is live! 🎉

Example:

# Add a new agent called "summarizer_agent"
uv run agentcore configure --entrypoint summarizer_agent.py

# Commit and push to deploy
git add .bedrock_agentcore.yaml
git commit -m "Add summarizer agent"
git push origin main

# ✅ Agent automatically deployed to AWS Bedrock

• agent.py - Agent entrypoint
• .bedrock_agentcore.yaml - Agent configuration
• Dockerfile - Container definition
• .github/workflows/deploy-agent.yml - Deployment workflow

Every deployment is tracked and versioned automatically:

Each successful deployment creates timestamped artifacts:

• Naming: bedrock-agentcore-config-{agent-name}-{commit-sha}
• Retention: 90 days
• Purpose: Full audit trail of deployments and configuration changes

• Each git push to main creates a new agent version in AWS Bedrock
• Docker images are tagged with the commit SHA for traceability
• Deployment artifacts link code changes to production versions

This means you can:

• 📜 Review deployment history across all agents
• 🔍 Trace production issues back to specific commits
• ⏮️ Roll back to any previous configuration
• 📊 Audit who deployed what and when