Soham Ganatra, co-founder, Composio Image: Selvaprakash Lakshmanan for Forbes India
Even before the emergence of generative AI, software engineers were familiar with the challenges of getting disparate, and increasingly complex, software to talk to each other. Application programming interfaces (APIs) ushered in a big change in this regard.
At Sampark Inc, better known for its platform, Composio, co-founders Soham Ganatra and Karan Vaidya are dragging the concept to the AI era. They’re offering a developer-focussed integration platform built specifically for AI agents and large language models (LLMs)—enabling AI systems to talk to each other and the real world.
“Modern software today is like kind of compound products. They require data pulling in from different sources out there,” Ganatra tells Forbes India. “So, integrations are a necessary part of product building. It’s so integral that without it, the product actually can’t exist.”
That’s at the heart of what Composio is solving: Integrations can take up as much as a third or more of a developer’s time, while they’re among the more tedious aspects of building software.
Composio offers a platform where AI agents and LLMs connect to hundreds of tools, from resources such as GitHub to HubSpot, with just a line of code. It’s a ‘developer-first’ toolset that cuts integration time from months to days and improves reliability of AI agents from around 50 percent to as much as 90 percent, the entrepreneurs claim.
Ganatra and Vaidya, engineering graduates from IIT-Bombay, have known each other from their days together as physics Olympiad teammates in 2013. They founded Composio (Sampark) in 2023, based in San Francisco and Bengaluru.
Composio is what software product specialists such as Nate Jones call ‘middleware’. “The premise is simple: Let developers focus on what the AI should do, and let Composio handle how to connect to the apps and data that enable those actions,” Jones explains in a March 8 edition of his popular newsletter on Substack.
The platform is also an example of an emerging tech called model context protocol, according to Jones. The traditional path is daunting if one wants to build an AI assistant to handle mail, a platform like Slack, a customer management software and a database all at the same time. There are multiple tasks that are tedious and can take months, he writes.
“Composio acts as a unifying layer between AI agents (like LLMs) and the web of third-party services they need to use,” he explains. It offers features such as pre-built integrations, managed authentication, and better execution, which reduces problems such as LLM hallucinations.
“So for a new developer, it’s the effort of going live that differs. There’s easily a 95 percent reduction in their effort,” Ganatra says. “We have seen projects where they thought it would take them four to five months to completion. We have literally seen them go live in a week.”
“One of the biggest challenges in agentic AI is integration—AI agents need to seamlessly interact with enterprise applications, tools, and APIs without human intervention. That’s where Composio comes in,” Manav Garg, founding managing partner at Together, a Bengaluru VC fund, and an early backer of the startup, writes in a February post on LinkedIn.
“In many ways, Composio is to agentic AI what Twilio was to communication APIs: A foundational layer that will unlock an entire industry,” Garg adds. This is why over 14,500 developers, including teams from Meta, Salesforce, and Cisco, have adopted Composio’s framework, he says.
Today that number has more than doubled, according to Ganatra. “We are growing at more than 50 percent month-on-month in terms of just users,” he says. “In terms of revenue, we are in seven figures right now,”.
In March, Together teamed up with Lightspeed Venture Partners and Elevation Capital for a $24 million Series A investment in Composio.
Looking ahead, a top priority is finding the right people for the right roles. On the product front, “where we are going is trying to create a new layer for AI agents”, he says.
This means, as AI agents and their interactions with multiple software and services become more common, that layer must allow “any sort of intelligence to learn from experience, to create and to have intuition”. With more complicated operations, can the agent get better over time, he asks.