CodeGPT is intertwining human language with code 🌏

CV Deep Dive

Today, we’re talking with Alvaro Chavez, Co-Founder and CEO of CodeGPT.

CodeGPT is an Agentic platform designed to bridge the gap between human language and coding, making development teams generate higher quality code, simply and quickly. Founded by Alvaro Chavez and his co-founders, Danny and Gustavo, the company emerged from an experimental tool that went viral on the VSCode marketplace. Their mission is to revolutionize software development by making it accessible to everyone, regardless of their coding experience.

CodeGPT uses agent-based technology to streamline your development process, allowing users to upload private data as context in the form of text or a complete code repository to AI Agents that can be used in IDEs and across developer workflows.

Today, CodeGPT is used by hundreds of thousands of developers across 180 countries, including individuals at some of the world’s largest tech companies. The platform is designed to be flexible and accessible, catering to everyone from junior developers to seasoned engineers and even non-coders who are just starting. With features like autocomplete, Agent code review, index codebase graph, and Chat Agent code Experts, CodeGPT aims to redefine how software is created in the age of AI.

In this conversation, Alvaro takes us through the founding story of CodeGPT, the unique challenges of building an agent-based coding platform, and the company’s vision for the future of software development.

Let’s dive in ⚡️

Read time: 8 mins

Our Chat with Alvaro 💬

Alvaro - welcome to Cerebral Valley! First off, give us a bit about your background and what led you to found CodeGPT? 

Hey there! My name is Alvaro, and I’m the CEO and co-founder of CodeGPT. I’ve always been involved in tech—I started out in college studying computer science, but quickly switched over to business while staying close to the tech world. 

My last Corporate job was at an airline, where I worked in fuel efficiency and procurement, managing billions of dollars in jet fuel. But what I enjoyed most was the fuel efficiency work, which involved a data science team managing millions of data points per flight to improve fuel usage. That experience really sparked my interest in data & tech, so I decided to leave that job and dive into entrepreneurship, that was 2017.

I first got into prop tech, sold that business after a year, and then bootstrapped a dev shop with two co-founders. I sold most of my shares in that business two years ago and then started another proptech, we saw an opportunity and we take it which led me to move to the US. We raised around $100 million in equity and debt to acquire single-family rental homes. Then the market froze, and that’s when I noticed what Danny, my co-founder, and Gustavo were working on. The story is pretty wild—Danny had started experimenting with OpenAI’s API before ChatGPT even launched, back when it was still GPT-2.

When ChatGPT did launch, Danny saw an opportunity and, with Gustavo, created a tool to help them code faster and more efficiently. They listed it on the VS Code marketplace, and it quickly went viral. By January, Danny was posting on LinkedIn about having 100,000 users, then 200,000, and then 300,000. He even shared a video of his six-year-old daughter, Ayla, coding with AI. Danny had built an interface with voice and ears for the bot, and Ayla was using it to code. 

That’s when I realized Danny was onto something big. So I called him up and asked, “What exactly are you doing?” He said he wasn’t even sure; he just created this tool, and it took off. Someone from Norway had already offered him $500,000 to buy it. That’s when I said, “No, let’s build something bigger.” And that’s how we got started.

Give us a top level overview of CodeGPT - how would you describe the startup to those who are maybe less familiar with you? 

Our vision of the platform is all about bridging the gap between human language and coding with the help of AI Agents. In practical terms, what we’ve built is an agent-based platform to enhance developers’ workflows, Designed to help development teams generate higher-quality code, simply and quickly, with an AI that is context-aware You can upload text and full repositories of code as a context into our system, and then use those agents in the IDE or any developer workflow tools like Slack, Teams or Discord, through Chrome, or via an API.

We offer a range of features from autocomplete and chat interfaces to inline coding, commit review, and much more. Essentially, everything revolves around this agent that knows your context, which enhances and streamlines your development process. The Agents are private, or users can select one from the Marketplace.

We also provide our users access to the best of the LLMs: OpenAI's models, Anthropic, Mistral, Google, and even open-source models. Additionally, we offer functionality that allows developers to work locally with their own model or using Ollama.

Who are your users today? Who’s finding the most value in what you’re building with CodeGPT? 

We have users in over 180 countries, with several hundred thousand people using our platform worldwide. About 30% of our users are based in the U.S., but we have a presence in every corner of the globe. Our users range from individuals at major companies—including some of the biggest names out there—to developers in Europe, Latin America, India, and beyond. The platform is built to cater to a wide range of users, from junior developers to senior engineers, managers, and even people who don’t know how to code but want to start. Our solution is flexible and accessible to everyone.

We want to impact millions of people, making coding easier for everyone. That's why we decided that our users can use the platform for free every day.

There are a number of teams working at the intersection of AI and code. What sets CodeGPT apart from the others in the space? 

Our AI Agent-centric and LLM agnostic approach is unique. The CodeGPT platform is built around AI agents, making it more adaptable and context-aware compared to traditional code completion tools. In the near future, developers will be using this agents to do much more than today. This approach allow us to have better performance and more precise answers beyond a chat interface.

That's why we’re really good at adding context to the AI Agents—not just text and code, but we’ve created our own solution to add an entire code repository as context, reducing the hallucinations by an order of magnitude, which is a major issue when you’re coding and using RAG for the same purpose. And we’ve solved that problem. Our AI Agents, when a repository of code is added to them as context, know your specific code, the names of the variables, and functions, which is a game changer for those working on big projects.

We like to think that our vision and great team set us apart. We are building the future of how software is created. In the next five to ten years, the landscape is going to be totally different. It’s likely that we’ll be writing software in human languages—English, Spanish, Portuguese, Chinese, Arabic, you name it. The number of people who can create software won’t just be limited to the 25 to 28 million developers out there now; it’s going to expand to anyone who wants to create, this is what drives our execution and what shapes our roadmap.

That’s why we built this platform based on agents. Our agent-based platform is unique in that you can integrate these agents everywhere. We’re continuously adding tools to these agents. Right now, people might see it as just an interface, but it’s capable of so much more. You can execute tasks—like clicking a button on our platform and having an agent review your commit, find any issues, and send you an email with all the problems it found.

We’ve also integrated code interpreter tools, so executing code in the future will be seamless. Everything is based on this agents that integrate into and enhance your existing workflows with your unique and private knowledge. 

With CodeGPT.co you have:

An AI-Assisted Coding with Context Awareness: CodeGPT's AI agents have a holistic view of the project, allowing for more accurate and contextually relevant code suggestions. This goes beyond simple auto-complete, as the AI understands the broader codebase and project structure.

Onboarding and Knowledge Sharing: CodeGPT can be used to create onboarding agents that make specific knowledge widely available within an organization. This can significantly reduce the time it takes for new developers to become productive on a project.

API and Framework Expertise: The AI agents can act as experts in specific APIs, languages, or frameworks, reducing the learning curve for developers working with new technologies.

Cross-Platform Integration: CodeGPT's agents are portable and can be accessed from various platforms including IDEs (VS Code, IntelliJ), communication tools (Slack, Discord), and even directly in GitHub for pull requests and issues.

How do you plan on CodeGPT progressing over the next 6-12 months? Anything specific on your roadmap that new or existing customers should be excited for? 

We think agents are going to be the key moving forward. You need to have an agent that can ingest context from various sources, and that’s something we're improving on continuously. We already have a way to ingest full repositories of code, and it works really well. But the next step is to ingest code from databases and basically context from everywhere. Once you have all that knowledge in the agent, it’s the closest thing you can get to a fine-tuned model. While it’s not exactly fine-tuning, it serves as a pretty solid proxy.

After that, what's coming next, in our view, is executing tasks with these agents. They’re going to be capable of doing more than just code completion—they’ll be able to create software, iterate on it, and even execute it automatically, moving toward what we envision as an auto-healing code roadmap. 

The next step after that is agents communicating with each other. I kind of imagine it like a brain, where each neuron is an agent, and they’re all communicating to arrive at a solution. The answer they provide could end up being a fully functional platform.

How do you measure the impact that CodeGPT is having on your key customers? Any customer success stories that you’d like to share?  

We know the impact that CodeGPT creates from the feedback we receive from our users, which is generally very positive. Internally, we measure KPIs such as auto-complete acceptance rate and interactions. We estimate that developers using CodeGPT save between 30% to 40% of their time.

One story we like is about Asfi, who we met in San Francisco. He used to be a banker, but now he’s diving into coding and building his own platform using our Python agent. He wasn’t very familiar with Python before, but thanks to Code GPT, he’s been able to learn and start creating his solution. We have several stories like these, which are really encouraging for us.

What’s the hardest technical challenge you’ve had to face whilst building CodeGPT to where it is today? 

One of the most significant challenges we’ve been focusing on is how to handle an entire repository of code in an efficient way so that an agent can truly understand and work with it.

The problem is that using a full repository of code, especially a large one, poses a unique challenge. We initially tried Retrieval-Augmented Generation (RAG), but it didn’t quite work out for us. RAG is probabilistic, so while it’s useful, it doesn’t work well when you need exact knowledge of a codebase, particularly with things like variable names where you can’t afford to change even a letter. You need precise knowledge of the code.

We’ve been working hard on solving this issue, and we’ve developed our own algorithm and process to work with full code repositories. This feature is currently in beta, with around 400 users testing it. The agent is be able to connect to your private or public code repository in Gitlab, Github, or Bitbucket, understand it fully, and provide the exact code you need, including all connections and dependencies. It’s really something we’re proud of. 

When you're working with a full repository of code, this feature changes everything. It's like having a ChatGPT specifically for your code repository. Eventually, you'll be able to query it in a way that goes beyond just understanding the code—you could ask it to fix bugs, and it wouldbe able to do that.

Just yesterday, we tried something new and were able to detect and fix vulnerabilities in a code repository. We still need to test more, but these kinds of features are coming soon. If anyone reading this wants to try this, just send me an email.

Lastly, tell us a little bit about the team and culture at CodeGPT. How big is the company now, and what do you look for in prospective team members that are joining?

We're always on the lookout for good talent, and we're currently hiring developers. Our team is remote, spread out across different locations, and we're all about finding talent wherever it is.

Every week, we have an all-hands meeting with the entire company to see what everyone is working on. Back in April, one of our non-developer team members develop something similar to the artifacts of Cloud 3.5 (before it was even launched). It was impressive, and that's the kind of culture we foster—we encourage everyone to be creative, propose ideas, and make changes whenever they see value. If it's valuable and align with our vision, let's do it. Internally, we call it the "Viking mode"—just do it and execute, and not be afraid to fail just keep trying and improving. It doesn't need to be perfect at the start; the key is to get it done, and then improve. We even have an internal joke about who is the "Viking of the Week," recognizing someone who's created something new or made a significant contribution. 

Anything else you’d like people to know about the work you’re doing at CodeGPT? 

Our platform has always been agnostic to any AI model, and I think that’s crucial for developers. We’ve made it so they can use any model they want, and we’ve maintained that flexibility. If a new model hits the market today, it’s likely available on our platform the same day, ready for everyone to use. The great thing is, you don’t have to choose a single model for the entire platform—you can pick and choose based on the feature or agent. For instance, you might prefer Anthropic for one task, Mistral for another, OpenAI for something else, and Gemini for yet another and with CodeGPT you have all of them in one subscription

You can even use open-source models if that’s what works for you. I think that’s really important for the developer community, and it’s something we’re committed to improving. Our goal today is simple: talk to users, understand what adds value to them and what doesn’t, and keep improving the platform to be the best. That’s the goal.

Conclusion

To stay up to date on the latest with CodeGPT, learn more about them here.  

Read our past few Deep Dives below:

• Zilliz Cloud is powering vector search for the enterprise 🔋

• Kolena aims to redefine quality standards in ML 🎛

• Pinecone Serverless is turbo-charging your time-to-prod ⚡️

• Cleanlab is automating data curation at scale 🌐

If you would like us to ‘Deep Dive’ a founder, team or product launch, please reply to this email ([email protected]) or DM us on Twitter or LinkedIn.

Join Slack | All Events | Jobs