17 KiB
🔮 AI Agents
Welcome to our list of popular AI autonomous agents. Your favorite one is missing? Add them via pull request. Discussion and feedback appreciated ❤️
Open Source Projects
AgentGPT
A browser-based implementation of AutoGPT, accessible via a no-code platform
How it works
- A no-code platform
- Assigning a goal to the agent, witnessing its thinking process, and formulation of an execution plan and taking actions accordingly
Features
- Uses OpenAI functions
- Supports gpt-3.5-16k, pinecone and pg_vector databases
Links
Stack
- Frontend: NextJS + Typescript
- Backend: FastAPI + Python
- DB: MySQL through docker with the option of running SQLite locally
Links
AI Legion
Similar in spirit to AutoGPT and Baby AGI, but written in TypeScript
AIrplane
Note: Airplane isn't AI agent. It's a company with a product for building internal tools. They're building an AI assistant here https://www.airplane.dev/autopilot. This is the link I'd be using. It's closed source.
Turning APIs, SQL queries, and scripts into apps for the entire team
How it works
- A developer-centric approach to building internal UIs and workflows
Features
- Airplane lets you turn SQL queries, JavaScript/Python code, HTTP requests, etc into tasks
- Allows to run tasks through a no-code dashboard
- Tasks for customer support, on-call runbooks, and scheduled tasks
Links
Aomni
Note: Closed source.
An AI agent specifically designed for research
How it works
- Breaks down a high level research question into a step-by-step plan, and executes it
- Diverse tools, including a full web browser
- Can access internet information without the need for an API
- "We don't generate content using AI, as it can be unreliable. Instead, we extract relevant information from trusted sources, cluster and process it into a user-friendly format."
- AI-powered query planner intelligently routes and executes requests, ensuring correctness and diverse source selection
Links
AutoGPT
- A lot like BabyAGI combined with LangChain tools.
- Can execute many commands such as Google Search, browse websites, write to files, and execute Python files
- GitHub
BabyAGI
How it works
- Creates tasks based on the result of previous tasks and a predefined objective.
- The script then uses OpenAI's NLP capabilities to create new tasks based on the objective
- Leverages OpenAI's GPT-4, pinecone vector search, and LangChainAI framework
- Default model is OpenAI GPT3-turbo
- The system maintains a task list for managing and prioritizing tasks
- It autonomously creates new tasks based on completed results and reprioritizes the task list accordingly, showcasing the adaptability of AI-powered language models
###Links
BabyCatAGI
BabyCatAGI is a mod of BabyBeeAGI, which is a mod of OG BabyAGI
How it works
- Just 300 lines of code
- This was built as a continued iteration on the original BabyAGI code in a lightweight way. Differences to BabyAGI include the following:
- Task Creation Agent runs once
- Execution Agent loops through tasks
- Task dependencies for pulling relevant results
- Two tools: search_tool and text_completion
- “Mini-agent” as tool
- Search tool combines search, scrape, chunking, and extraction.
- Results combined to create summary_report
How to use
- Fork this into a private Repl
- Add your OpenAI API Key (required) and SerpAPI Key (optional)
- Update the OBJECTIVE variable
- Press "Run" at the top.
Links
BitBuilder
Closed source. Link to their website, not GitHub marketplace - https://www.bitbuilder.ai/.
"Virtual intern for developers", generates code via PRs raised against your repository
How it works
- Create an Issue
- Approve the Implementation Plan
- Review the Pull Request
Features
- Writing code
- Reviewing changes
- Addressing comments
- Answering questions
<<<<<<< HEAD
Virtual interns for developers.
Task them with writing code, reviewing changes, addressing comments, answering questions, and more. You gotta check our
You gotta check our
ec08b882a6@bitbuilder_ai , it raises pull requests against your repository. In fact the backend is built with @tiangolo ’s FastAPI so I can personally attest to its ability to create routers, endpoints, and wire it all together.
We’ve made more than a pretty demo. @bitbuilder_ai will create Pull Requests from Issues. And review your code.
BitBuilder Junior
- BitBuilder Junior is an intern-level code generation tool that lives in your GitHub repository. It follows instructions to implement unamibigous code changes by changing multiple files, writing tests for generated code, and matching the style, framework, and libraries you're already using.
Statistics
- 990+ REPOSITORIES use BitBuilder, including private and organization repos
Links
- Slack
- [Installation(https://docs.bitbuilder.ai/install)]
Bloop
Just make sure to communicate that it's a code search engine that uses an AI agent.
Butternut AI
Closed source.
Camel
Communicative Agents for “Mind” Exploration of Large Scale Language Model Society
It's more like an agent architecture than an agent itself. Just make sure you communicate this.
- AI user agent: give instructions to the AI assistant with the goal of completing the task.
- AI assistant agent: follow AI user’s instructions and respond with solutions to the task
Links
- Paper - CAMEL: Communicative Agents for “Mind” Exploration of Large Scale Language Model Society
- Colab demo
- GitHub
- Hugging face datasets
- Slack
- Authors: Guohao Li∗ Hasan Abed Al Kader Hammoud* Hani Itani* Dmitrii Khizbullin, Bernard Ghanem
Cognosys
Closed source.
Web based version of AutoGPT/babyAGI
- Friendly UI for building AI agents
- Author: Sully Omarr
- Twitter: https://twitter.com/SullyOmarr
Databerry
Factory
Closed source.
Fixie
Closed source.
Friday
Closed source.
A developer assistant able to make whole nodejs project with unlimited prompts
How it works
- Provides a core prompt for building the foundation of your application
- Allows you to add unlimited sections, each of which is a prompt representing a specific part of your app
Features
- Friday utilizes GPT-4 for AI assistance, but it has been tested and optimized with GPT-4-32k for improved speed and better results.
- It requires 2 small requests for your app's base and 1 request per section you provide.
- Friday employs esbuild behind the scenes for every app created by it.
Links
- Author: Amirreza Salimi
GitWit
GPT Engineer
Specify your project, and the AI agent constructs the entire code base.
GPT Engineer is made to be easy to adapt, extend, and make your agent learn how you want your code to look. It generates an entire codebase based on a prompt.
Model: GPT 4 Specify what you want it to build, the AI asks for clarification, and then builds it.
Features
- You can specify the "identity" of the AI agent by editing the files in the identity folder.
Editing the identity, and evolving the main_prompt, is currently how you make the agent remember things between projects.
Each step in steps.py will have its communication history with GPT4 stored in the logs folder, and can be rerun with scripts/rerun_edited_message_logs.py.
Links
- Author: Anton Osika
- Twitter review by @Attack
Grit
Closed source.
HayStack Agent
This is not an agent but SDK for creating agents. Like LangChain or LlamaIndex.
Heymoon.ai
Closed source Personal assistant for life: to keep you on top of your calendar, tasks and information
Features
- Personal assistant for life: to keep you on top of your calendar, tasks and information. Was at Llama event demo
Hyperwrite
Closed source. Not sure if agent-like. Probably more like generative AI.
Your personal AI writing assistant
Jarvis
LastMile AI
I spoke with them, they're building an SDK that should be something like "Langchain for enterprises". Not an agent
Loop GPT
LocalGPT
Inspired by privateGPT
How it works
- Most of the description on readme is inspired by the original privateGPT
- Model: Vicuna-7B
- Using InstructorEmbeddings
- Both Embeddings as well as LLM will run on GPU. It also has CPU support if you do not have a GPU
- Built with Langchain
Features
- Ask questions to your documents without an internet connection, using the power of LLMs. 100% private, no data leaves your execution environment at any point. You can ingest documents and ask questions without an internet connection!
Links
Minion AI
Closed source. Made by creator GitHub Copilot.
MultiOn
Closed source.
Naut ai
Closed source. I'll be talking to a founder next Thursday.
Pezzo
Saga
Maybe link directly to their AI product? https://saga.so/ai It's not an agent yet though, I think. They're moving in the agent direction.
Second
Closed source.
Smol developer
Superagent
Not a single agent. You can create agents without coding with this tool.
Sweep
Closed source.
Teenage AGI
Can recall infinite memory, THINKS before it speaks, and doesn't lose memory after being shutting down
How it works
- Model: GPT-4
- Language: Python
- Uses OpenAI and Pinecone to give memory to an AI agent and also allows it to "think" before making an action (outputting text)
- Also, just by shutting down the AI, it doesn't forget its memories since it lives on Pinecone and its memory_counter saves the index that it's on
Here is what happens every time the AI is queried by the user:
- AI vectorizes the query and stores it in a Pinecone Vector Database
- AI looks inside its memory and finds memories and past queries that are relevant to the current query
- AI thinks about what action to take
- AI stores the thought from Step 3
- Based on the thought from Step 3 and relevant memories from Step 2, AI generates an output
- AI stores the current query and its answer in its Pinecone vector database memory
Links
- Created by @sean_pixel !
“Westworld” simulation
Westworld is a multi-agent simulation library, its goal to simulate and optimize systems and environments with multiple agents interacting.
- Researchers from Stanford and Google created an interactive sandbox env with 25 Gen AI agents can simulate human behavior
- They walk in the park, join for coffee at a cafe, and share news with colleagues. They demonstrated surprisingly good social
- Westworld's inspiration is drawn from Unity software and Unity ML Agents, adapted in Python
Underlying paper - Generative Agents
- A paper simulating interactions between tens of agents
- Presenting an architecture that extends a language model to store and synthesize the agent's experiences, enabling dynamic behavior planning in an interactive sandbox environment with generative agents
Links
Languages
- The library is available on PyPi via pip install westworld
- Javascript version (being developed)
Current features
- Easy creation of Grid and non-grid environments
- Objects (Agents, Obstacles, Collectibles, Triggers)
- Subclassing of different objects to create custom objects
- Spawner to generate objects randomly in the environment
- Basic rigid body system for all objects
- Simple agent behaviors (pathfinding, wandering, random walk, fleeing, vision range)
- Automatic maze generation
- Layer integration to convert image to obstacle and snap it to a grid
- Sample simulations and sample agents for classic simulations
- Simulation visualization, replay and export (gif or video)
Yourgoal
Open source
Swift implementation of BabyAGI.
Companies building agents
Hex Magic
👋 Wanna discuss AI agents and more?
💪 Contributors 💪
Made with contrib.rocks.