KINTO Tech Blog
Generative AI

Agent Store v1.0 Launch: An In-House Platform to Streamline Internal Agent Development

Cover Image for Agent Store v1.0 Launch: An In-House Platform to Streamline Internal Agent Development

Introduction

Hi, I'm Alex from the AI First Group. As AI technology rapidly advances, demand for agent development is growing fast. However, getting started with efficient agent development can be challenging. Information about the latest tools is fragmented, and development resources are limited. To tackle this, we've launched Agent Store v1.0, a platform to share internally developed AI agents and centralize our technical know-how.

Purpose of Agent Store

The Agent Store has two primary purposes:

  1. Boost in-house agent development efficiency
    • Shorten development cycle by reusing existing agents
    • Enables rapid builds using templates
  2. Build technical know-how
    • Centralize management of in-house Agent-related technologies
    • Facilitate <i>yokoten</i> (horizontal deployment) of best practices

Usage Patterns and Target Users

Agent Store is a platform where employees can freely develop, share, download, and reuse AI agents. For reusability, we envision it like an app store—users can download agents and deploy them in their own environments.

The Agent Store is made up of the following components:

  • Github repository for sharing internally developed agents
  • How CI/CD works for agent development

Agent StoreのGithubリポジトリ Github repository for Agent Store

Usage Format

Agent Store v1.0 primarily supports AWS Bedrock agents.

We designed the agent’s CI/CD process with an Infrastructure as Code (IaC) approach in mind. Agents shared through the Agent Store are stored as SAM templates in its GitHub repository, making them ready for deployment via CloudFormation.

Users looking to reuse an agent can download its SAM template and deploy it to their environment using CloudFormation. Deployment is automated using Github Actions.

Those who want to build new agents can also jumpstart development using the blank SAM templates available in the Agent Store.

Target Users (v1.0)

  • Engineers looking to start building agents using AWS Bedrock
  • Engineers wanting to share Bedrock-built agents within their organization
  • Engineers looking to discover and utilize existing agents

** Expected Use Cases**

Agent Store is designed for the following scenarios:

  1. Creating new AI agents
    • Speed up development using Agent Store's CI/CD flow
  2. Sharing AI agents
    • Share custom-built agents across your company
  3. Reusing shared AI agents
    • Equip existing products with AI agents
    • App development to streamline internal operations
    • Avoid building from scratch and boost development efficiency
  4. ** AI agent PoC**
    • Deploy existing agents to quickly conduct PoC
    • Shortening the time required for effect verification
  5. ** Acquire know-how for developing AI agents**
    • Learn best practices by looking at similar cases
    • Lowering technical barriers

Agent Development, Sharing, and Reuse Flow Using Agent Store

  1. ** Flow for developing a new agent** The flow begins by retrieving a blank SAM template from the Agent Store repository, filling it out, and deploying it.

    • For more information about SAM, see here.
    • If you're using an agent's Action Group, make sure to fill out and deploy the templates in this order: ①Lambda, ②Agent.  Flow for developing a new agent
  2. ** Flow for sharing agents ** Prepare a SAM template for the agent you developed and submit a pull request to the Agent Store repo. The agent reviewer will check the content and merge it if there are no problems. Agent sharing flow

  3. Flow for reusing a shared agent The overall process is similar to developing a new agent, but it begins by retrieving the SAM template of a shared agent from the Agent Store. After making any needed changes or additions, you can move on to deployment. Agent reuse flow

Agent Architecture

Here's the architecture of the agent. After deployment, the Bedrock agent can invoke and execute Lambda functions configured as Action Groups when needed. Stack management is handled through CloudFormation. Agent Architecture

You can also build multi-agent collaboration,where multiple agents work together. Multi-agent architecture

Future Development Plans

To boost adoption of the Agent Store, we're planning a variety of internal study sessions, workshops, and hackathons.

Currently, v1.0 is designed for engineers with AWS Bedrock experience (Type A), but we plan to expand our target users in stages. | User type | Description | Support status || ---- | ---- | ---- || Engineer type A | Engineers experienced with AWS Bedrock | Supported in v1.0 || Engineer type B | Engineers interested in agent platforms other than AWS Bedrock | In planning || Non-engineers / Beginners | Employees with no coding or development experience | In planning |

Summary

Agent Store v1.0 is a platform designed to streamline agent development and foster knowledge sharing. Currently, it's available to AWS Bedrock users, but we plan to expand support to a broader user base and integrate with a wider range of agent frameworks. To make the most of our in-house AI development resources and accelerate innovation, we're committed to actively expanding and evolving Agent Store.

Facebook

関連記事 | Related Posts

We are hiring!

生成AIエンジニア/AIファーストG/東京・名古屋・大阪・福岡

AIファーストGについて生成AIの活用を通じて、KINTO及びKINTOテクノロジーズへ事業貢献することをミッションに2024年1月に新設されたプロジェクトチームです。生成AI技術は生まれて日が浅く、その技術を業務活用する仕事には定説がありません。

【クラウドプラットフォームエンジニア】プラットフォームG/東京・大阪・福岡

プラットフォームグループについてAWS を中心とするインフラ上で稼働するアプリケーション運用改善のサポートを担当しています。

イベント情報