Ex-Twitter CEO Parag Agrawal unveils new AI startup and research API
Former Twitter CEO Parag Agrawal’s new venture Parallel Web Systems has launched the Deep Research API for real-time web research.
Former Twitter CEO Parag Agrawal has launched a new artificial intelligence startup, Parallel Web Systems, unveiling its flagship product called Deep Research API.
The platform is designed to enable AI agents to conduct advanced web research in real time, with the goal of surpassing human capabilities in specific research tasks.
A new venture after Twitter
Agrawal, who served as Twitter’s CEO before Elon Musk’s takeover in 2022, has returned to the technology spotlight with his Palo Alto-based company.
Parallel Web Systems has raised around $30 million in funding from investors including Khosla Ventures, Index Ventures, and First Round Capital, according to reports.
The company announced that the Deep Research API is already powering “millions of research tasks daily,” supporting activities such as document discovery, code debugging, and workflow automation.
Agrawal stated that the API is designed to deliver results with “exceeding human-level accuracy” in certain specialised contexts.
Features of the Deep Research API
Parallel Web Systems has built eight distinct AI research engines within the API, each optimised for different computational tasks.
These engines handle use cases such as long-form synthesis, cross-disciplinary analysis, and knowledge retrieval. The product is positioned as a tool for developers and organisations seeking to integrate advanced research capabilities into their applications.
According to company claims, the Deep Research API outperforms leading models, including OpenAI’s GPT-5, on select benchmarks related to web-based research.
While independent evaluations are still limited, early demonstrations suggest its potential in handling complex, multi-step queries.
Industry relevance and applications
The API addresses a growing need for AI systems capable of accessing and analysing live information, beyond the static training data on which most models operate.
For enterprises, this could enable use cases ranging from market intelligence and legal research to scientific analysis.
Developers are expected to use the system for building autonomous research agents and improving decision-support tools.


