Micro Agent

Description

Micro Agent is an AI-powered tool designed to assist developers by writing and refining code. Its core functionality revolves around generating code that satisfies predefined test cases, thereby automating the iterative process of debugging and improvement. This approach aims to address the common challenge of LLMs producing code that requires extensive manual correction.

The tool operates by first generating a test case and then iteratively producing code until all specified tests pass. This test-driven development (TDD) methodology is central to Micro Agent's design, ensuring that the generated code is not only functional but also robust and verifiable. The project emphasizes a 'micro' agent philosophy, meaning it is intentionally focused on a single, well-defined task: writing tests and then generating code that passes those tests. This contrasts with more general-purpose AI agents that can sometimes lead to compounding errors.

Micro Agent supports integration with several popular LLM providers, including Claude, OpenAI, and Ollama, as well as any OpenAI-compatible services like Groq. Users can configure their API keys and preferred models directly through the command-line interface. The tool offers two primary modes of operation: unit test matching and an experimental visual matching mode. In unit test matching, users provide a test script that Micro Agent executes after each code generation attempt. The visual matching feature, which is still under development, allows the agent to attempt to match a given design by comparing rendered code output against a screenshot.

Installation is straightforward via npm, requiring Node.js v18 or later. Getting started can be as simple as running `micro-agent` in interactive mode, which guides the user through the code generation process. For more advanced use cases, users can specify file paths, test scripts, prompt files, and even visual targets. The project also includes options for configuring the maximum number of runs and managing settings through a dedicated `config` command or environment variables. Contributions are welcomed, with a focus on improving the agent's capabilities and user experience.

Micro Agent's Core Features

• AI-powered code generation and iteration

• Test-driven development approach

• Supports multiple LLM providers (Claude, OpenAI, Ollama, Groq)

• Interactive mode for guided code generation

• Unit test matching for code verification

• Experimental visual matching for design adherence

• Configurable API keys and models

• Customizable maximum run attempts

• Integration with Visual Copilot for Figma-to-code conversion

• Command-line interface for configuration and execution

Getting Started with Micro Agent

1. 1

   Install: Run `npm install -g @builder.io/micro-agent`

2. 2

   Configure API Key: Use `micro-agent config set OPENAI_KEY=<your token>` or `micro-agent config set ANTHROPIC_KEY=<your token>`

3. 3

   Set Model: Use `micro-agent config set MODEL=<model name>` (e.g., `gpt-4o`, `claude-3-5-sonnet-20241022`)

4. 4

   Run Interactively: Execute `micro-agent` and follow prompts

5. 5

   Run with Tests: Use `micro-agent <file path> -t "<test script>"`

6. 6

   Run with Visual Matching: Use `micro-agent <file path> --visual <url>` (requires Anthropic API key)

Micro Agent's Use Cases

• Automated Code Refinement
• Test-Driven Development Assistance
• Rapid Prototyping
• Reducing Manual Debugging
• Design-to-Code Implementation
• LLM Code Generation Improvement
• Developer Productivity Boost

FAQ from Micro Agent

• What is Micro Agent and what does it do?

  Micro Agent is an AI tool that writes and iteratively fixes code. It generates code based on test cases and refines it until all tests pass, aiming to improve code quality and reduce manual effort.

• Which LLM providers does Micro Agent support?

  Micro Agent supports Claude, OpenAI, Ollama, and any OpenAI-compatible providers such as Groq.

• How does Micro Agent ensure code quality?

  It employs a test-driven development approach, generating code that must pass specified unit tests. It iterates on the code until all test cases are satisfied.

• What is the 'micro' aspect of Micro Agent?

  The 'micro' designation signifies its focused purpose: to excel at writing tests and then producing code that passes those tests, avoiding the complexity and potential errors of general-purpose coding agents.

• Can Micro Agent help with visual design implementation?

  Yes, Micro Agent has an experimental visual matching feature that attempts to match rendered code output to a provided design screenshot. It also integrates with Visual Copilot for Figma-to-code conversion.

• What are the prerequisites for installing Micro Agent?

  Micro Agent requires Node.js v18 or later to be installed on your system.

• How do I configure API keys for Micro Agent?

  You can configure API keys using the command line, for example: `micro-agent config set OPENAI_KEY=<your token>` or `micro-agent config set ANTHROPIC_KEY=<your token>`.

• Can I specify which AI model Micro Agent uses?

  Yes, you can set the preferred model using `micro-agent config set MODEL=<model name>`, such as `gpt-4o` or `claude-3-5-sonnet-20241022`.

• What is the 'unit test matching' mode?

  In this mode, you provide a test script that Micro Agent runs after each code generation attempt. The agent continues to generate code until the script executes successfully.

• How can I contribute to the Micro Agent project?

  Contributions are welcome. You can submit Pull Requests and check the `./CONTRIBUTING.md` file for guidelines. Issues labeled 'good first issue' are also available for new contributors.

• What is the default number of runs Micro Agent attempts?

  By default, Micro Agent attempts 10 runs. This can be changed using the `-m` flag, for example: `micro-agent ./file-to-edit.ts -m 20`.

• Does Micro Agent handle complex multi-file operations?

  No, Micro Agent is designed as a 'micro' agent and does not attempt to install modules, read/write multiple files, or perform other complex operations that could lead to significant errors.