Skip to content

README_EN.md

Latest commit

 

History

History
189 lines (128 loc) · 9.5 KB

README_EN.md

File metadata and controls

189 lines (128 loc) · 9.5 KB

AWS TechBot

License: MIT-0 IaC: CloudFormation Powered by AgentCore Built with Strands

English | 简体中文

An AI-powered AWS technical assistant built with Strands Agents SDK and Amazon Bedrock AgentCore.

Tools are managed through AgentCore Gateway — the Agent automatically discovers all available tools at startup.

Use Cases

  • Documentation Lookup — @mention the bot in Feishu/Lark to query AWS service docs, best practices, and architectural guidance
  • Configuration & Tutorials — Look up configuration steps, tutorials, and sample code for AWS services
  • Cost Estimation — Real-time pricing queries for both AWS Global and China regions
  • Troubleshooting — Quickly find troubleshooting guides, quota limits, error codes, and operational SOPs

Deployment

Prerequisites: Create a Feishu/Lark App

Before deploying, create an app on the Feishu Open Platform to get the App ID and App Secret:

  1. Go to Feishu Open Platform and create an enterprise app
  2. Go to Credentials & Basic Info, copy App ID and App Secret
  3. Enable Bot capability
  4. Configure permissions
  5. Copy Verification Code

See the Feishu Bot Setup Guide (Steps 1-5) for detailed instructions

One-Click CloudFormation Deployment

Region Deploy
US West (Oregon) Launch Stack

Click the button and fill in the parameters. The stack creates:

  • AgentCore Gateway — Unified MCP tool endpoint + Cognito auth
  • Gateway Targets — 4 Lambdas (Global Knowledge, China Knowledge, Pricing, Customer Stories)
  • AgentCore Runtime — Runs the TechBot container
  • AgentCore Memory (optional) — Multi-turn conversation memory
  • API Gateway — /chat POST endpoint for Feishu webhook
  • Handler Lambda — Receives Feishu events, filters @all, async invokes worker
  • Worker Lambda — Calls AgentCore, updates Feishu card with response

Parameters:

Parameter Required Description
Model ID Pre-filled GLM-5 (Best Performance) Nova 2 Lite, GLM-5, MiniMax M2.5 (only Nova 2 Lite supports image input)
Enable Memory Pre-filled true true for multi-turn memory, false for stateless
Memory Expiry Days Pre-filled 30 days Memory expiry in days (7-365)
Feishu App ID Required Feishu app credentials
Feishu App Secret Required Feishu app credentials
Feishu Verification Token Required Feishu webhook verification token (Open Platform → Events & Callbacks → Encryption Strategy)
  • Leave all other options (Tags, Permissions, Stack failure options, etc.) as default.
  • Check ✅ I acknowledge that AWS CloudFormation might create IAM resources with custom names at the bottom.
  • Click Create stack and wait for CREATE_COMPLETE (~5 minutes).

Post-Deployment: Complete Feishu Configuration

After the stack is deployed, copy FeishuEventSubscriptionUrl from Outputs and complete the Feishu setup:

  1. Configure Event Subscription — Paste the URL into Feishu Open Platform → Events & Callbacks → Request URL
  2. Add Events — Add im.message.receive_v1 event
  3. Publish App — Create a version and submit for approval
  4. Add to Group — Add the bot to a Feishu group

See the Feishu Bot Setup Guide (Steps 6-9) for detailed instructions

Done!

See the Usage Guide for features and examples

Architecture

diagram

Model Pricing

Model Input (per 1M tokens) Output (per 1M tokens) Image Input Characters
Nova 2 Lite $0.33 $2.75 Yes Support images input
MiniMax M2.5 $0.30 $1.20 No Best cost performance
GLM-5 (Zhipu AI) $1.00 $3.20 No Best answer quality (Recommended)

Model Selection Guide:

  • GLM-5 (Recommended) — Most detailed and clear answers, strongest tool-calling ability, best overall experience
  • MiniMax M2.5 — Good answer quality at the lowest price, ideal for cost-sensitive use cases
  • Nova 2 Lite — The only model supporting image input, but may occasionally fail or return incomplete answers on complex queries (e.g., multi-step pricing lookups)

Cost Estimation

All charges are pay-as-you-go. No usage = no cost. No upfront fees, no minimum spend.

Based on real-world testing (documentation queries, pricing lookups, China region service checks, customer story searches). Estimated for 300 questions/month (~10/day).

Model Invocation Cost

Pricing queries cost more per call due to multi-step tool usage. Numbers below are averaged across query types.

Model Per Query (avg) Monthly (300 queries)
MiniMax M2.5 ~$0.012 ~$3.7
Nova 2 Lite ~$0.015 ~$4.4
GLM-5 ~$0.041 ~$12.3

AgentCore Infrastructure Cost

Each question triggers 1 Runtime invocation and ~5 Gateway API calls on average.

Service Description Monthly
Runtime CPU + Memory, consumption-based < $3
Gateway ~5 API calls per question < $0.01
Memory (optional) Multi-turn conversation memory < $0.5
Lambda / API Gateway Within free tier

Total Monthly Cost

Model Model Cost Infrastructure Total
MiniMax M2.5 ~$3.7 < $4 < $8
Nova 2 Lite ~$4.4 < $4 < $9
GLM-5 ~$12.3 < $4 < $17

All services are pay-as-you-go — no cost when idle. Actual costs vary based on query complexity (number of tool calls, response time) and Memory settings. Only Nova 2 Lite supports image input.

Customization

After deployment, you can customize the solution to fit your needs.

Modify Agent Behavior

Edit MAIN_SYSTEM_PROMPT in main.py to adjust the agent's response style, scope, or instructions. Rebuild and update the Runtime after changes.

Add New Tools

  1. Create a new Lambda function implementing your tool logic
  2. In the AgentCore Gateways console, select your Gateway → Targets → Add. Provide the Lambda ARN and tool schema definition
  3. The agent will automatically discover the new tool on next startup — no agent code changes needed

Update Agent Image

After modifying code, build a new image, push to ECR, and update the Runtime:

# Build ARM64 image
docker buildx build --platform linux/arm64 -t techbot:latest .

# Push to ECR
AWS_REGION=us-west-2
ACCOUNT_ID=$(aws sts get-caller-identity --query Account --output text)
aws ecr get-login-password --region $AWS_REGION | docker login --username AWS --password-stdin $ACCOUNT_ID.dkr.ecr.$AWS_REGION.amazonaws.com
docker tag techbot:latest $ACCOUNT_ID.dkr.ecr.$AWS_REGION.amazonaws.com/techbot-repo:latest
docker push $ACCOUNT_ID.dkr.ecr.$AWS_REGION.amazonaws.com/techbot-repo:latest

# Update AgentCore Runtime (DEFAULT endpoint auto-points to latest version)
aws bedrock-agentcore-control update-agent-runtime \
    --agent-runtime-id "your-runtime-id" \
    --agent-runtime-artifact '{"containerConfiguration":{"containerUri":"'$ACCOUNT_ID'.dkr.ecr.'$AWS_REGION'.amazonaws.com/techbot-repo:latest"}}' \
    --network-configuration '{"networkMode":"PUBLIC"}'

The Runtime ID can be found in CloudFormation Outputs.

Cleanup

To remove all deployed resources, go to the AWS CloudFormation console, select the TechBot stack, and click Delete. All associated resources (Runtime, Gateway, Lambda, Cognito, Memory, etc.) will be automatically removed.

Disclaimer

This is sample code for demonstration purposes only. You should work with your security and legal teams to meet your organizational security, regulatory, and compliance requirements before deployment. Deploying this solution may incur AWS charges.

Security

Security is a shared responsibility between AWS and the customer. This sample deploys resources within your AWS account — you are responsible for securing your account, managing IAM permissions, and configuring services according to your organization's requirements. AWS is responsible for the security of the underlying cloud infrastructure. For more information, see the AWS Shared Responsibility Model.

See CONTRIBUTING for reporting security issues.

License

This project is licensed under the MIT-0 License. See the LICENSE file.