Outerport is a platform that transforms diagrams, drawings, webpages, documents, CAD files, and other visual or textual assets into precise text representations that can be ingested by large language models (LLMs) or accessed via APIs for use by AI agents. By leveraging advanced computer vision and vision-language models (VLMs), Outerport enables enterprises to conduct search and build AI agents that go beyond text, parsing complex visual information into actionable data and code.
The platform stands out by converting visual diagrams into code that can not only reproduce the original but also expose APIs, making visual data as accessible to AI as text. Outerport's agentic search system improves upon traditional vector RAG (Retrieval-Augmented Generation) and long-context methods, delivering faster and more accurate retrieval through innovations like KV cache management and chain-of-memory reasoning. This enables complex queries requiring reasoning or comprehensive answers, and does so with 10-20x greater efficiency compared to legacy approaches.
How Was Outerport Started?
Outerport was co-founded by Towaki Takikawa (CEO) and Allen Wang (CTO). Towaki Takikawa is a former research scientist at NVIDIA, known for his work on high-performance systems for machine learning and graphics, with research published in top conferences and widely cited. Allen Wang previously held tech lead roles at startups like Tome and Embark Trucks, focusing on LLM applications and computer vision, and has additional experience from LinkedIn, Meta, and Borealis AI. Both founders are alumni of the University of Waterloo, with strong backgrounds in computer science and machine learning.
What Problem Does Outerport Solve?
Outerport addresses the challenge of making non-text data—such as diagrams, CAD files, and complex documents—usable by AI systems and searchable by enterprises. Traditional AI search and agent systems are often limited to text, but Outerport's platform enables robust, multimodal data handling. This expands the scope of enterprise search and task automation to include visual knowledge, aiding industries where diagrams and drawings are critical.
Who Uses Outerport?
While specific customer names are not publicly available, Outerport is designed for enterprises and organizations that need to extract, search, and automate workflows using both text and complex visual data, such as engineering, architecture, manufacturing, and knowledge management teams.
How Does Outerport's Technology Work?
- Converts visual assets (diagrams, CAD files, documents) into structured text/code using computer vision and VLMs
- Exposes APIs for AI agents to interact directly with processed visual data
- Utilizes agentic search with advanced techniques like chain-of-memory reasoning and KV cache management for accurate, efficient retrieval from large document sets
Leadership Team
- Towaki Takikawa, Co-Founder / CEO: Research scientist background at NVIDIA, recognized in machine learning and graphics
- Allen Wang, Co-Founder / CTO: Experienced in LLM applications, computer vision, and large-scale data systems from Tome, Embark Trucks, LinkedIn, and Meta
Recent News
Outerport recently published a blog post titled "How agentic search helps AI understand long documents", detailing how their chain-of-memory reasoning allows AI agents to retrieve information reliably from extremely long contexts without hallucination. The post, co-authored by both founders, highlights their ongoing innovation in the space.
Use PromptLoop to Uncover Company Data
Looking for more company insights like this? PromptLoop helps you go deeper, providing unique data points and analysis on companies like Outerport and many others. Automate your research and find the information that matters most. Discover how PromptLoop can accelerate your market intelligence. Get A Free Demo to learn more.