No-Code Personal Agent Built on GPT-4.1 and the Realtime API

Super Agent, launched by Genspark, is a no-code personal AI agent that performs real-world tasks—such as making phone calls, creating presentations, and producing videos—using only simple prompts, without any coding. The service recorded explosive growth, reaching $36 million (ARR) just 45 days after launch.

At the core of the system are OpenAI's multimodal models.

• GPT-4.1: Handles research and structured output based on a 1 million-token context window and strong reasoning capabilities.
• GPT-image-1: Supports image generation for presentations and video production.
• Optimization techniques: Automatic prompt caching reduces latency and API costs, improving the efficiency of multi-step workflows.

In particular, the 'Call For Me' feature, which leverages the OpenAI Realtime API, performs tasks such as making reservations or changing schedules through natural voice conversations. It operates on a dual-layer structure where the Realtime API manages the real-time conversation while a shadow model monitors and guides the interaction through a message queue, ensuring highly fluent conversations.

Genspark has pivoted from an existing AI search engine to agent-centric AI, demonstrating overwhelming execution capability by rolling out 8 major agent features in 70 days with a team of just 20 people.