2025 will be the year of AI apps, changing how we build and use technology.
From smarter tools to powerful frameworks, the tech world is changing fast and AI is shaking things up in a big way.
Today, we will look at the top 22 tools and open source frameworks for building AI apps, plus a bonus list at the end.
Let's jump in.
🎯 What are AI Agent frameworks?
In this list, there are a lot of AI agent frameworks so let's take a moment to understand what it means.
AI agents are like really smart assistants. You just tell them what you need, and they figure out how to get it done!!
The LLM acts as the brain of the system. When an AI has to communicate with the outside world, obtain data, or carry out particular tasks, it can utilize tools, which are external resources or APIs.
They can plan, make decisions, and even get better over time. It’s more like having a helper that works independently without needing you to guide it every step of the way!
AI agents can be mainly of two types:
⚡ Reactive: Respond to immediate inputs from their environment.
⚡ Proactive Agents: Plan ahead to achieve long-term goals.
If you're looking to learn more, What is an Agent Framework by Lyzr is a great place to start. It covers Architecture, the need for AI agents, key components and what factors to consider while choosing the right framework.
1. CopilotKit - 10x easier to build AI agents.
You will agree that it's tough to add AI features in React, that's where Copilot helps you as a framework for building custom AI Copilots.
You can build in-app AI chatbots, and AI Agents with simple components provided by Copilotkit which is at least 10x easier than building it from scratch.
✅ Grounded in real time user-specific context.
✅ Easily integrate LangChain & LangGraph agents into your copilot.
They provide built-in (fully-customizable) Copilot-native UX components like <CopilotKit />, <CopilotPopup />, <CopilotSidebar />, and more.
Get started with the following npm command.
npm i @copilotkit/react-core @copilotkit/react-ui
CopilotPopup is a convenience wrapper for CopilotChat that lives at the same level as your main content in the view hierarchy. It provides a floating chat interface that can be toggled on and off. Here's how you can use it:
import { CopilotPopup } from "@copilotkit/react-ui";
export function YourApp() {
return (
<>
<YourMainContent />
<CopilotPopup
instructions={"You are assisting the user as best as you can. Answer in the best way possible given the data you have."}
labels={{
title: "Popup Assistant",
initial: "Need any help?",
}}
/>
</>
);
}
You can read the docs and check the demo video.
You can integrate Vercel AI SDK, OpenAI APIs, Langchain, and other LLM providers with ease.
They have also recently released CoAgents, the framework for embedding Agents in your application.
CopilotKit has 15k+ stars on GitHub.
2. LangChain - build context-aware reasoning AI apps.
LangChain is great for creating sequences of prompts and managing interactions with AI models. It has a large ecosystem with lots of pre-built integrations, which makes it easier to run things quickly.
Developers can design powerful AI agents with complicated reasoning, task execution while interacting with external data sources and APIs.
It's one of the most popular LLM frameworks with a massive community. Under the hood:
uses LangGraph to build stateful agents with first-class streaming and human-in-the-loop support.
uses LangSmith to inspect, monitor and evaluate your applications, so that you can continuously optimize and deploy with confidence.
You can read the docs and learn more about LangChain.
There is a very interesting video on YouTube by IBM, definitely recommended!
LangChain has 98k stars on GitHub and is used by 160k+ developers.
3. Aider - AI pair programming in your terminal.
Aider is like having a smart pair programmer in your terminal.
It lets you pair programs with LLMs, to edit code in your local git repository. Start a new project or work with an existing code base. Aider works best with Claude 3.5 Sonnet, DeepSeek V3, o1 & GPT-4o and can connect to almost any LLM.
✅ You can add images to the chat.
✅ You can add URLs to the chat.
✅ You can code with your voice.
✅ It can edit multiple files at once for complex requests.
It uses a map of your entire git repo, which helps it work well in larger codebases. I've attached a small demo!
If you are not aware, Aider has one of the top scores on the SWE Bench. SWE Bench is a challenging software engineering benchmark where aider solved real GitHub issues from popular open source projects like django, scikitlearn, matplotlib, etc.
You can read the docs.
You can watch this demo!
Aider has 24.9k stars on GitHub.
4. Bolt - prompt to build.
Bolt.new is an AI web development agent that's pretty wild. You just describe what you want and you've got a full-stack app right in your browser. You don't even need to set up a local environment!
It's actually built on top of the Bolt open source repo and powered by the StackBlitz WebContainer API. Most people are not even aware that the repo they provide just contains the UI interface and the server components for Bolt. You can read more if you're curious.
It uses Anthropic's Claude AI (which is considered very good) and also has some cool features like real-time preview as you work, automated debugging and one-click deployment to places like Netlify.
You don't even need to leave your browser to get from idea to live app.
If you're looking to build something but are confused about whether it can even build a decent app? Just watch this.
Bolt has 11.8k stars on GitHub.
5. AgentOps - Observability and DevTool platform for AI Agents.
AgentOps helps developers build, evaluate and monitor AI agents, from prototype to production. They provide a Python SDK for AI agent monitoring, LLM cost tracking, benchmarking, and more.
Some of the things you can do:
✅ Fine-tune specialized LLMs up to 25x cheaper on saved completions.
✅ You can rewind and replay agent runs with point in time precision.
✅ Keep data of logs, errors and prompt injection attacks.
✅ Visually track events such as LLM calls, tools and multi-agent interactions.
You can start by using the command: pip install agentops.
You can integrate with most LLMs and agent frameworks including CrewAI, Langchain, Autogen and CamelAI.
You can read the docs.
If you are looking for a decent tutorial, you can watch this!
AgentOps has 2.6k stars on GitHub.
6. LangGraph - building stateful, multi-actor agents with graphs.
LangGraph is an advanced library built on top of LangChain, designed to improve your Large Language Model (LLM) apps by introducing cyclic computational capabilities.
While LangChain allows the creation of Directed Acyclic Graphs (DAGs) for linear workflows, LangGraph takes this a step further by including the addition of cycles, which are essential for developing complex, agent-like behaviors.
These behaviors allow LLMs to continuously loop through a process, dynamically deciding what action to take next based on evolving conditions.
Most people often think that LangChain is very similar to LangGraph even though it's not. You can watch this video to understand the differences.
You can read the docs.
If you're looking for a guide, How to Build AI Agents with LangGraph: A Step-by-Step Guide is a really good article with concepts explained in a simple way.
LangGraph has 7.9k stars on GitHub.
7. E2B - secure open source cloud runtime for AI apps.
E2B is an open-source infrastructure that allows you to run AI-generated code in secure isolated sandboxes in the cloud. You can think of it as a small computer for the AI model. You can run many sandboxes at once.
They provide JavaScript SDK and Python SDK to start and control those sandboxes.
Some of the usual use cases for E2B are AI data analysis or visualization, running AI-generated code of various languages, playground for coding agents, environment for codegen evals, or running full AI-generated apps like in Fragments.
You can install JavaScript SDK as follows.
npm i @e2b/code-interpreter
Then set up your E2B API key and execute code with code interpreter inside Sandbox as follows.
import { Sandbox } from '@e2b/code-interpreter'
const sandbox = await Sandbox.create()
await sandbox.runCode('x = 1')
const execution = await sandbox.runCode('x+=1; x')
console.log(execution.text) // outputs 2
You can read the docs and also check how to Analyze data with AI using E2B.
It helps you run AI related code securely in the cloud, making it good enough for building and testing AI apps in 2025.
E2B has 7.3k stars on GitHub.
8. CrewAI - role-playing, autonomous AI agents.
CrewAI is a framework for orchestrating role-playing AI agents. It allows developers to create a crew of AI agents, each with specific roles and responsibilities, to work together on complex tasks.
It kind of imitates human organizational structure, building upon the idea of multi-agent systems. Here's how it works:
- the crew organizes the overall operation
- AI agents work on their specialized tasks
- the process ensures smooth collaboration
- tasks get completed to achieve the goal
You can read the docs and about the concepts involved.
There is a short course by AndrewNg (Deeplearning) on how to build Multi AI Agent Systems with crewAI.
It includes code samples, 19 video lessons and is recommended since courses by Deeplearning are always great. Just do this!
CrewAI has 24k stars on GitHub.
9. Better Auth - most comprehensive authentication framework for TypeScript.
Authentication in the TypeScript ecosystem is a half-solved problem. Other open-source libraries often require a lot of additional code for anything beyond basic authentication. Rather than just pushing third-party services, we need something better.
If you're building a next level AI app, you would need a simple yet proper authentication system. You might have heard about Clerk, Next Auth but there is one other which is my favorite (yet many people don't know about it).
Better Auth is a framework-agnostic authentication (and authorization) library. It provides a comprehensive set of features out of the box and includes a plugin ecosystem that simplifies adding advanced functionalities with minimal code in a short amount of time.
You can start using it with the following command.
npm install better-auth
This is how you can use it.
// auth.ts
export const auth = betterAuth({
database: new Pool({
connectionString: DATABASE_URL,
}),
emailAndPassword: {
enabled: true,
},
plugins: [
organization(),
twoFactor(),
]
})
// client.ts
const client = createAuthClient({
plugins: [passkeyClient()]
});
Social Sign-On options include Apple, Discord, Facebook, GitHub, Google, Microsoft, Twitch, Twitter (X), Dropbox, LinkedIn, GitLab, Reddit and Spotify.
I really loved the part about plugins. For instance, if we want to enable anonymous authentication, we can simply do this.
import { betterAuth } from "better-auth"
import { anonymous } from "better-auth/plugins"
export const auth = betterAuth({
// ... other config options
plugins: [
anonymous()
]
})
There are many more plugins like Google One-Tap, PassKey (using cryptographic key pairs), or even Multi Session that allow users to maintain multiple active sessions across different accounts in the same browser.
You can read the docs and the live demo at demo.better-auth.com.
Better Auth has 6k stars on GitHub.
10. Tavily - connect your LLM to the Web.
Tavily helps your AI apps get real-time, accurate information from the web, works well with AI models like LLMs and tools that use retrieved data effectively (RAG).
Tavily provides a very good Search API and eliminates the need for manual SERP scraping by providing fast, reliable access to web data with content snippets improved for AI processing.
What is the use case?
Let's say you are building an AI chatbot, Tavily will help the bot to fetch up-to-date, contextually relevant information directly from the web. This improves the bot's ability to answer complex queries accurately and be more reliable to users.
You can start by using this command.
pip install tavily-python
This is how you can generate a precise and fact-based context for your RAG application in one line of code.
from tavily import TavilyClient
# Step 1. Instantiating your TavilyClient
tavily_client = TavilyClient(api_key="tvly-YOUR_API_KEY")
# Step 2. Executing a context search query
context = tavily_client.get_search_context(query="What happened during the Burning Man floods?")
# Step 3. That's it! You now have a context string that you can feed directly into your RAG Application
print(context)
You can read the docs.
You can see this video about the open research agent-native app (ANA) which uses Copilotkit, Tavily and LangChain!
Their JavaScript SDK is in beta right now.
Their Python wrapper for Tavily search API has 376 stars on GitHub.
11. Microsoft AutoGen - programming framework for agentic AI.
AutoGen is a framework for creating multi-agent AI apps. It can do a bunch of things simultaneously and even handle live data streams.
Developed by Microsoft Research (as you could guess by the name), the ecosystem provides everything you need to create AI agents, especially multi-agent workflows, frameworks, developer tools and apps.
For instance, this is how you can create an assistant agent using OpenAI's GPT-4o model.
import asyncio
from autogen_agentchat.agents import AssistantAgent
from autogen_ext.models.openai import OpenAIChatCompletionClient
async def main() -> None:
agent = AssistantAgent("assistant", OpenAIChatCompletionClient(model="gpt-4o"))
print(agent.run(task="Say 'Hello World!'"))
asyncio.run(main())
It also supports two essential developer tools:
-
AutoGen Studio: provides a no-code GUI for building multi-agent applications. -
AutoGen Bench: provides a benchmarking suite for evaluating agent performance.
You can read the docs.
If you're looking for a nice tutorial, check this!
Autogen has 37k stars on GitHub.
12. Rasa - automate text and voice-based conversations.
Rasa is a conversational AI platform that enables teams to build chatbots, voice assistants and other automated conversation systems.
It provides the tools and infrastructure to develop, train, and deploy contextual and intelligent virtual assistants that can understand natural language.
With Rasa, you can build contextual assistants on Facebook Messenger, Slack, Google Hangouts, Webex Teams, Microsoft Bot Framework, Rocket.Chat, Mattermost, Telegram, Twilio. And your own custom conversational channels/voice assistants such as Alexa Skills, Google Home Actions.
They actually provide two main solutions:
Rasa Pro: python framework for building scalable, dynamic conversational AI assistants that integrate LLMs.Rasa Studio: no-code graphical user interface that helps business users to collaboratively build, review and improve conversational user journeys at scale.
You can read the docs which contain both the products and how to use them.
One of the co-founders shared detailed insights on building bots for better contextual conversations in this Medium article: A New Approach to Conversational Software.
Rasa has 19k stars on GitHub.
13. Supabase - perfect database with useful AI features.
Supabase has turned into a proper backend solution with some cool AI features.
It includes vector similarity search for handling AI recommendations or semantic searches, built-in support for storing and managing embeddings and even a plain English to SQL (yeah it's cool).
It also has real-time subscriptions that are ideal for building live AI-driven apps like chat systems or collaborative tools. Plus, their edge functions let you run AI computations close to users, reducing latency for tasks like real-time inference.
Get started with the following npm command (Next.js).
npx create-next-app -e with-supabase
You can read the docs.
If you're new to Supabase, watch this video to understand it in under 100 seconds!
You can build a crazy fast application with Auth, realtime, Edge functions, storage and many more. Supabase almost covers it all!
The best part is that they provide a lot of AI templates and examples such as Vector search with Next.js and OpenAI.
Supabase has 76k stars on GitHub.
14. AutoGPT - more exciting than ChatGPT.
At the core of AutoGPT lies its primary project, a semi-autonomous agent driven by large language models (LLMs), designed to perform any tasks for you.
It's a platform that allows you to create, deploy, and manage continuous AI agents that automate complex workflows.
You can read the docs and the quickstart guide on how to build Agent with AutoGPT.
If you're interested, watch this video by FireShip that explains AutoGPT!
💡 What can you build with this?
For example, you can build an agent that has the ability to browse Reddit which will then identify trending topics and automatically create short-form videos based on the content.
It's a very powerful combination and opens a lot of opportunities. Even if you don't know much about AI, you can try AutoGPT to understand how you can save time and build cool stuff.
AutoGPT has 170k stars on GitHub.
15. TanStack - server-state management with powerful libraries.
TanStack provides powerful libraries for managing server-state in JavaScript and TypeScript apps, making it easier to handle data fetching, caching, synchronization and more.
TanStack Query is one of those libraries, it auto refreshes data, has smart caching that actually works, handles real-time updates without headaches, does optimistic updates and has awesome devtools (that you cannot live without).
Some of the other libraries are also decent to use.
AI apps often need to fetch and manage large datasets, whether for training models, processing inputs or interacting with APIs. Using TanStack Query could help you optimize the data flow and reduce the complexity of managing asynchronous data (even if it's a little bit).
You should also check out TanStack Start which is a new full-stack react framework. To be honest, I never really heard about it.
TanStack has multiple repos but the query one has 43k stars on GitHub.
16. Lyzr - build reliable, safe and responsible AI agents.
This is the only framework on the list which is not open source. It's built with a strong focus on safety and responsible AI.
It makes sure that AI agents work ethically, with features to handle privacy, fairness, and transparency right out of the box.
Even though it’s not open source, it makes up for it with great support and tools that are polished for professional use.
They also provide Agent studio where you can build an agent in a few minutes. You will also find community agents and a lot of other important stuff.
✅ Create, manage and deploy agents with a simple, low-code interface.
✅ Integrate workflows with existing tools.
There is also a YouTube tutorial on how to build your first AI agent in under 60 seconds with Lyzr!
17. Phidata - build multi-modal Agents with memory, knowledge, tools and reasoning.
Phidata is a framework for building multi-modal agents and workflows.
You can build teams of agents that can work together to solve problems. Plus, you can chat with your agents using a beautiful Agent UI.
You can install using this command.
pip install -U phidata
It focuses on minimal code, so for instance, you can create a web search agent in just 10 lines of code.
from phi.agent import Agent
from phi.model.openai import OpenAIChat
from phi.tools.duckduckgo import DuckDuckGo
web_agent = Agent(
name="Web Agent",
model=OpenAIChat(id="gpt-4o"),
tools=[DuckDuckGo()],
instructions=["Always include sources"],
show_tool_calls=True,
markdown=True,
)
web_agent.print_response("Tell me about OpenAI Sora?", stream=True)
You can read more on the docs and test things out on their playground accessible at phidata.app/playground.
They provide templates and also example apps that can help you get started faster.
Phidata has 17k stars on GitHub.
18. BotPress - hub to build and deploy AI Agents.
Botpress is another platform for building AI agents. Its visual workflow design, extensive AI integrations and multi-channel support make it useful for automating customer interactions and workflows.
In other words, it's great for building custom chatbots without needing a lot of coding skills. It's easy to use, works well for businesses and helps create chatbots that can handle big tasks and connect with other tools easily.
There are lots of features such as:
✅ You get a drag-and-drop interface for no-code bot creation.
✅ It will automatically sync with your databases & systems.
✅ You can deploy bots on websites, WhatsApp, Slack, Instagram, Telegram, Messenger and more platforms.
You can read the docs and check out their academy course on how to build AI agents. They also provide pre-built templates!
You should watch this tutorial showing how to build custom trained AI chatbots using Botpress.
Botpress has 13k stars on GitHub.
19. Storybook - UI development, testing, and documentation made easy.
Storybook is a frontend workshop for building UI components and pages in isolation. It helps in UI development, testing, and documentation.
For AI apps, it could be useful for creating interactive UIs that use AI in some form such as dashboards, data visualizations or something that interacts with AI models or APIs. You can build and showcase components in isolation, without needing to load the entire app context.
You can start using it with this command.
npx storybook@next init
This is how you can create a simple component for your project.
import type { Meta, StoryObj } from '@storybook/react';
import { YourComponent } from './YourComponent';
//👇 This default export determines where your story goes in the story list
const meta: Meta<typeof YourComponent> = {
component: YourComponent,
};
export default meta;
type Story = StoryObj<typeof YourComponent>;
export const FirstStory: Story = {
args: {
//👇 The args you need here will depend on your component
},
};
You can read the docs.
These days UIs are painful to debug because they are entangled in business logic, interactive states and app context.
Storybook provides an isolated iframe to render components without interference from app business logic and context. That helps you focus development on each variation of a component, even the hard-to-reach edge cases.
Confused? Watch this to understand it in 100 seconds!
Storybook has 85k stars on GitHub.
20. Visual Copilot - convert your figma designs to code.
A lot of international teams use Figma for their UX designs and you know, how hard it can be to make a pixel-perfect user interface.
If there is an existing solution, why not give it a try?
Visual Copilot is a powerful Figma plugin that turns your Figma designs into clean, ready-to-use code. There's no setup needed and it can generate code for almost any framework such as React, Next.js, Vue, Svelte, Angular, Swift, Flutter, Kotlin, React Native, HTML and more.
You can even choose your preferred styling library such as Tailwind CSS, CSS Modules, Emotion, Styled Components or Styled JSX.
Best part? Generated code is automatically responsive, even when your designs don't use auto layout. Wow!
I personally haven't tried it because I used AdobeXD (prefer offline) and just don't design things before coding. It's still worth the try!
Watch the official tutorial.
As per the official Figma page, it has 832k users.
21. BaseAI - first web AI framework for building Serverless AI agents.
BaseAI is the first web AI framework for building Serverless AI agents with Node.js and TypeScript.
🎯 What the heck is serverless agents with memory?
In simple words, it's an AI agent that runs serverless without you having to manage the infrastructure. It solves the universal limitations of LLMs in that they don't have your customized data. That's where memory comes in.
You can build serverless and composable AI agents with memory and tools. It allows you to develop AI agent pipes on your local machine with integrated agentic tools and memory (RAG).
✅ local developer experience.
✅ zero bloatware no boilerplate code.
✅ deploy to Serverless AI cloud with the npx baseai deploy command
To create a new BaseAI project, you can use the following command.
npx baseai@latest init
Then you will have to add API keys, create an agent, integrate pipe and finally run it. Just follow the guidelines in the docs.
You can read the docs and you will also find about the project structure, CLI and quickstart guides.
You can check this official demo and learn BaseAI from the learn section in the docs, they've taken examples of creating a summarizer pipe, creating a weather tool and more.
Check how to build a Next.js app with BaseAI.
BaseAI has 783 stars on GitHub. It's very new but has a lot of potential with an awesome use case.
22. Cursor - the perfect AI code editor.
This list would be incomplete without Cursor.
I have used VSCode for years and avoided AI editors due to a perception that AI-written code would take longer to debug. I assumed that it would just cancel the productivity but I gradually realized how wrong I was.
In my experience of using it, Cursor actually understands your project, unlike other editors who just say it for the sake of saying it.
It gets your coding style, knows your project structure and even picks up on your team's best practices. It's just that good.
It's a fork of VS Code so you can just import the settings, themes, keybindings and extensions to maintain the same experience.
There are many more features such as Tab, Composer, Chat, Terminal, ⌘ K and more!
If you're curious to know more features, models and terminologies like context or AI review, you can explore on the official docs.
When I was starting, I just read the blog The Ultimate Introduction to Cursor for Developers by the Builder.io team. This is a complete guide and is highly recommended!
Other alternatives are Continue, Zed, Void, PearAI, WindSurf.
Cursor is not open source but their support repo has 26.8k stars on GitHub.
There are more amazing AI agent frameworks you can take a look at:
LlamaIndex (GPT Index) - to structure and query external data with LLMs. Great for creating smarter AI agents that understand your data better.
Graph Reader - reads and analyzes data from graphs, helping AI agents process relationships and patterns in a better way.
OpenAI Swarm - focuses on collaboration between multiple AI agents to handle complex tasks together.
Vertex AI - google platform for building, deploying and scaling machine learning models, including tools to integrate AI agents easily into apps.
Promptflow - to track, compare and optimize prompts for building reliable AI interactions.
Flowise - open source no-code interface for LangChain, making agent-building easier without programming.
Ema - framework for managing and deploying AI agents in a better way. Good for simple and fast use cases.
AgentForce - helps deploy and orchestrate powerful AI agents in real-world situations.
Glean - focused on enterprise AI, Glean helps organizations build AI agents that can find and organize internal data efficiently for teams.
Cohere - language AI platform for building agents with powerful natural language understanding. Perfect for creating conversational agents or text-driven workflows.
Microsoft Semantic Kernel - combines AI with planning and memory, ideal for building agents with long-term context.
LangFlow - GUI for LangChain that simplifies building AI workflows. Perfect for those who prefer drag-and-drop over coding.
I tried to cover unique and something useful in the list.
Whew! I know it's a lot but these projects would help you build your dream AI app in 2025.
I hope you found something useful here.
Have a great day! Till next time.
| If you loved this, please follow for more :) Thank you for reading, Anmol 🥰 |
  
|
|---|
Follow CopilotKit for more content like this.
Top comments (17)
Nice list! I'm thinking of a Tavily + CrewAI based agent to search, crawl information from websites and then do some function calling.
Thanks for reading Saurabh! 🙌 CrewAI is good, especially because you can assign pre-built tools to each agent.
There are too many combinations that could be useful depending on the use case. You can also check out: agent.ai/ (marketplace for AI agents)!
Sure!
Copilotkit is cool...
Even I wrote about it in one of my blog: dev.to/rohan_sharma/fly-with-ai-co... (not promoting!!)
I also wrote one article on co-agents
Its awesome
Thanks for reading, Rohan! It's okay to promote man :)
Even though many people may not like it but I always appreciate it since it builds trust and shows expertise. I hope others will check it out. 🔥
Excellent!
My tops are Langchain & CopilotKit
Once you have cursor, everything else is possible!
Yeah, I always loved using VSCode and only decided to try Cursor because of the hype on twitter. I never intended to use it as my default editor until I actually tried it.
Cursor just made building so much easier! 🔥
This is great!
Thanks for reading, Nathan! 🙌
If you plan to self host your Next.js applications, do checkout this article dev.to/simplr_sh/hosting-your-next...
Loved this curation. Super handy
You can also add Potpie in the list. An amazing platform to build AI Agents for complex codebases
github.com/potpie-ai/potpie
Amazing list for ai resources!
Thanks for reading buddy! 🙌 The list is definitely long and I've tried my best to include projects that cover most cases. There will always be some exceptions.
Nice list!
Good!!