Report Abuse

Basic Information

MiniAGI is a compact autonomous agent implementation that runs locally and interacts with OpenAI models (GPT-3.5-Turbo and GPT-4). The repository provides a runnable Python agent (miniagi.py) that accepts a natural-language objective and attempts to achieve it by using a small set of tools, short-term memory with summarization, chain-of-thought reasoning, inner monologue and optional self-criticism. Installation instructions and a minimal environment setup are included; Python 3.10+ is required and dependencies are installed via requirements.txt. The agent can execute Python and shell commands, perform web searches, read and write files, and interact with APIs when credentials are provided. The README includes multiple worked examples demonstrating tasks like drawing images, creating websites and games, scraping and summarizing content, generating PDFs, performing basic analyses, and automating multi-step workflows. A safety and privacy advisory warns about potentially harmful suggestions and use of third-party APIs.

Links

Categorization

App Details

Features
The project centers on a minimal but capable autonomous agent design. It ships with a robust prompt and a small tool set including execute_python, execute_shell, web_search and read_file as illustrated in examples. The agent implements short-term memory with summarization to retain context across steps, and supports chain-of-thought style reasoning, inner monologue and an optional critic mechanism that can be enabled via ENABLE_CRITIC in the environment to improve accuracy at the cost of extra API calls. Usage is simple: run python miniagi.py <objective>. The README documents many example flows showing how the agent composes commands, runs code, installs packages, serves a local webserver, scrapes sites, creates images and files, and asks the user for credentials when necessary. The repo includes an .env example and basic installation steps using pip and requirements.txt.
Use Cases
MiniAGI is useful for prototyping and running an autonomous task-oriented assistant that can carry out sequences of commands to achieve user-specified objectives. It automates multi-step workflows such as programmatically generating images, building and serving a website, scraping and summarizing content into PDFs, creating simple games, performing data analysis from CSV files, and orchestrating API-backed actions when credentials are available. The agent can reduce manual scripting by combining web search, code execution and file operations into a single loop driven by a natural language objective. It is also a learning tool for understanding agent prompts, short-term memory summarization and critic-based verification. The README notes privacy and safety caveats and shows how the agent requests user approval for sensitive actions like sending emails or using external API keys.

Please fill the required fields*