Genspark Super Agent: No-Code Autonomous Agents powered by GPT-4.1 and Realtime API

Genspark has released Super Agent, a no-code platform that enables users to automate real-world tasks—such as making phone calls, generating presentations, and creating short videos—using simple prompts. This release marks a pivot for Genspark from an AI search engine to an agentic AI platform, leveraging OpenAI's multimodal models and Realtime API to move from information retrieval to outcome-based automation.

Technical Architecture and Model Integration

Super Agent is built as an orchestration layer that manages nine specialized large language models and over 80 integrated tools. The system dynamically assigns tasks to the most appropriate component based on the user's request.

OpenAI Model Implementation

OpenAI models provide the core intelligence for the platform's workflows:

  • GPT-4.1: Used for research and the generation of structured outputs. It features a 1M-token context window, which allows the agent to process extensive documents without truncation, and improved instruction following.
  • Strict JSON Output: This feature ensures that responses are reliably structured for downstream tool integration.
  • Automatic Prompt Caching: This is utilized to reduce API costs and latency, particularly during complex, multi-step workflows.
  • GPT-image-1: This model is used via the OpenAI API for all image generation tasks, such as creating stylized cover images for presentations.

Multimodal Capabilities and Real-World Automation

Super Agent transitions AI from a chat interface to an actionable workspace through several key features:

Voice Automation via Realtime API

The "Call For Me" feature allows the agent to conduct natural, real-time phone conversations without scripts. This is powered by a dual-layer system:

  1. The Realtime API: Manages the live, speech-to-speech dialogue to ensure fluency.
  2. A Shadow Model: Monitors the interaction via a message queue to guide the conversation.

This architecture allows the Agent to handle ambiguous human responses and hold music effectively. In Japan, this has been notably used for handling resignation calls to employers.

Content Generation

Beyond voice, the system automates the creation of personalized media. For example, it can draft a pitch deck by writing the content, generating images with GPT-image-1, and compiling the final slides. Similarly, it can generate scene-by-scene scripts and images to assemble short-form videos for platforms like Instagram.

Business Impact and Growth Metrics

Since the launch of Super Agent in April 2025, Genspark has experienced rapid organic growth without paid marketing:

  • Revenue: Reached $36M in Annual Recurring Revenue (ARR) within 45 days.
  • Development Speed: Shipped eight major agent features in 70 days with a 20-person team.

Future Roadmap

Genspark plans to expand the capabilities of autonomous agents into new categories, including an AI browser that can act on web content in real-time and AI docs for the generation of format-rich documents.

Sources