Bedrock Engineer
Universal AI Agent using Amazon Bedrock, capable of customize to create/edit files, execute commands, search the web, use knowledge base, use multi-agents, generative images and more.
Last verified:
What is Bedrock Engineer?
Bedrock Engineer is an autonomous AI assistant for software development that uses Amazon Bedrock. It is a universal AI agent capable of creating and editing files, executing commands, searching the web, using knowledge bases, orchestrating multi-agent workflows, and generating images. Unlike editor-integrated assistants like Cline or Cursor, Bedrock Engineer has its own standalone native UI built with Electron, running on macOS, Windows, and Linux without requiring VS Code dependence.
Key features include Agent Chat with ReAct (Reasoning & Acting) methodology, where the AI autonomously reads source code, understands project structure, creates Mermaid.js diagrams, and implements code. It supports web grounding through Tavily Search and URL-specified searches, Agentic-RAG with Amazon Bedrock Knowledge Base forconforming to design guidelines and coding standards, multi-agent orchestration with sub-agents, and MCP (Model Context Protocol) server connectivity for external tools and resources. The tool includes built-in capabilities for file operations (reading, writing, moving, copying), folder creation, arbitrary command execution, image generation using Amazon Nova, Titan, and Stable Diffusion series models, and database/API connectivity.
Bedrock Engineer is designed for software developers, DevOps engineers, and AWS users who need an AI development assistant. It is particularly useful for teams wanting to customize AI agents to follow their coding conventions, architecture patterns, and design systems. The tool supports use cases beyond development, including creating specialized agents for infrastructure as code expertise, mobile app development, customer service scenarios, and educational content creation like kamishibai picture books.
The application includes security features like Application Guardrails for masking sensitive data, prompt caching to reduce costs, configurable context length settings for different task scales, a cost dashboard tracking token usage, and an Agent Directory for sharing developed agents and system prompts. Users can customize agents through system prompt auto-generation, create custom tools via Python API, and connect to MCP servers like AWS Documentation MCP, Knowledge Bases Retrieval MCP, AWS Cost Analysis MCP, and AWS CDK MCP.
Bedrock Engineer pricing
Pricing model: Freemium
Bedrock Engineer itself is free and open-source (GitHub repository aws-samples/bedrock-engineer). You only pay for Amazon Bedrock usage through your AWS account. Based on the presentation, cost estimates assume 100 sessions per month with Sonnet 3.7: approximately $120 for input tokens (400k tokens/session) and $30 for output tokens (20k tokens/session), totaling around $150/month. Actual costs are proportional to the number of sessions and token usage. A cost dashboard within the app calculates reference values based on session tokens and cached tokens, but users should refer to the AWS Management Console for accurate figures. Prompt caching is available to reduce costs.
Bedrock Engineer pros
- Standalone native app with own UI, no VS Code dependency required
- Supports macOS, Windows, and Linux through Electron framework
- Autonomous ReAct agent with reasoning and action history tracking
- Creates Mermaid.js diagrams to visualize software architecture
- Web grounding with Tavily Search for latest documentation research
- Agentic-RAG automatically optimizes queries against Knowledge Base
- Multi-agent orchestration supports sub-agents via Bedrock Agent or CLI
- MCP server connectivity for AWS Documentation, Cost Analysis, CDK, and more
- Agent customization with auto-generated system prompts for specialized tasks
- Built-in image generation using Amazon Nova, Titan, and Stable Diffusion
- Application Guardrails mask sensitive data before sending to LLM
- Prompt caching reduces costs for code analysis tasks
- Configurable context length (30-120 array length) for different task scales
- Cost dashboard tracks token usage and cached tokens per session
- Agent Directory enables sharing custom agents and system prompts
- Serverless architecture requires no AWS Lambda or CDK deployment
- Custom tool integration via Python API and arbitrary command execution
- Reads existing source code and understands project structure automatically
- Supports both Plan and Execute phases for granular control over AI behavior
Bedrock Engineer cons
- Windows users must build from source, no pre-built app available
- Requires AWS access key and secret access key for initial setup
- Need to manually request Bedrock model access in AWS Console first
- Only supports Amazon Bedrock, no other LLM providers available
- Security risk from MCP servers that might include malicious code
- Complex custom tools are difficult for AI to master properly
- Large tool execution results may exceed context window causing forgetting
- Costs can accumulate quickly with many sessions (estimated $150/month for 100 sessions)
- Context length over 120 risks excessive token use and task failure
- Code samples for Agentic-RAG must be prepared in advance in Knowledge Base
Frequently asked questions about Bedrock Engineer
What is Bedrock Engineer?
Bedrock Engineer is an autonomous AI assistant for software development that uses Amazon Bedrock. It is a universal AI agent capable of creating/editing files, executing commands, searching the web, using knowledge bases, orchestrating multi-agents, and generating images. Unlike editor-integrated assistants, it has its own standalone native UI built with Electron for macOS, Windows, and Linux.
How do I install Bedrock Engineer?
Mac users can download the app directly and click to install. Windows users need to build from source: clone the GitHub repository (aws-samples/bedrock-engineer), run 'npm install', then 'npm run build:win'. Mac users use 'npm run build:mac', Linux users use 'npm run build:linux'. After building, double-click the generated app to open it.
What AWS setup is required?
You need to set up AWS access key and secret access key from the settings screen after opening the app. Additionally, in the AWS Management Console, select Bedrock model access and request access to your desired models through 'Modify model access'. Only Amazon Bedrock is used; no AWS Lambda or CDK deployment is required.
What models can I use with Bedrock Engineer?
You can use any Amazon Bedrock model for which you have requested access. The default model is Claude 3 Haiku, but you can use Claude 3.5 Sonnet or other Bedrock models. You must request model access in the AWS Console first through the 'Model Access' section under Bedrock.
How does Agent Chat work?
Agent Chat uses ReAct (Reasoning & Acting) methodology. The AI interprets user queries, selects appropriate tools, executes them, and repeats this process until the objective is solved. It maintains reasoning and action history, adding each step to the conversation. The agent autonomously reads source code, creates folders, implements code, understands structure, creates Mermaid.js diagrams, and uses web search and RAG for optimal implementation plans.
What is Agentic-RAG and how is it different from traditional RAG?
Agentic-RAG automatically optimizes and executes multiple queries against the Knowledge Base without manual query refinement. Unlike traditional RAG requiring manual query refinement, the agent-based approach efficiently executes optimized queries. For example, it can search for 'Admin Interface: tables, forms, and data visualization' then 'Layout Components: ResponsiveLayout AppLayout Header' automatically to generate code conforming to design guidelines.
How do I customize agents for my team's conventions?
Use the Agent Customization feature: insert the agent's name and description, click the System Prompt auto-generation button, and the system produces a prompt incorporating tool specifications and contextual considerations. You can create specialized agents like 'Infrastructure as Code Expert' or 'Mobile App Development Expert' with specific capabilities, working guidelines, and best practices that follow your team's coding conventions and standards.
What MCP servers are available?
Available MCP servers include AWS Documentation MCP (read/search documentation, get recommendations), Knowledge Bases Retrieval MCP (discover knowledge bases, natural language query, data source filtering, reranking), AWS Cost Analysis MCP (analyze/visualize costs, query with natural language, optimization suggestions), and AWS CDK MCP (CDK guidance, CDK Nag integration, Solutions Constructs support, GenAI constructs). These enable inspecting AWS environments, error log analysis, GitHub issue research, Jira integration, browser testing, and database connections.
How do guardrails work for security?
Application Guardrails automatically mask sensitive data before transmission to LLMs. From the Settings screen, specify the Guardrail identifier and version to apply guardrails automatically via guardrailConfig in the Converse API. You can also use ApplyGuardrail API directly without going through the LLM, which is effective for verifying ToolUse execution results before including them in requests or when using non-Bedrock LLMs.
What context length should I use for different tasks?
Bedrock Engineer supports configurable context length (array length). Use 30 for very small-scale tasks (1-2 file changes), 60 for small-scale tasks (cause analysis with human assistance focusing on specific files), 90 for functional changes fitting in a single commit (investigate and implement specific functionality), and avoid 120 as it represents excessive token use where tasks should be broken down into smaller ones. The optimal length depends on information volume and token count per message element.