HeHo — Project Story

Inspiration

HeHo was inspired by a recurring problem in modern AI platforms: powerful models exist, but using them with your own data often means vendor lock-in, hidden storage, or complex infrastructure. I wanted to build something different—a platform where AI works directly with your database, stays fully open-source, and puts users in complete control. The core idea was simple: AI should adapt to your product, not the other way around.

What I Built

HeHo is a completely open-source, no-code AI chatbot builder that connects directly to a user’s own Supabase database. Chatbots can securely read database data and create new entries, all governed by fine-grained permissions. AI models are powered by OpenRouter, giving access to 35+ models without locking users into a single provider. Your data stays in your database—always.

How I Built It

HeHo is built using modern web technologies, with Supabase handling authentication, databases, and Row Level Security (RLS). OpenRouter provides flexible AI model routing. A permission layer ensures that every AI action is explicitly allowed. To reduce hallucinations, AI responses are constrained to verified database queries and controlled operations.

Inline logic follows the principle: \( \text{AI Capability} \le \text{Granted Permissions} \)

Challenges Faced

One of the biggest challenges was preventing AI hallucinations while still allowing useful autonomy. Another challenge was designing a system that feels powerful but remains no-code and safe. Ensuring everything stayed fully open-source without compromising security required careful architectural decisions.

What I Learned

I learned that trustworthy AI depends more on guardrails and ownership than raw intelligence. When AI understands real context and operates within strict boundaries, it becomes far more reliable and valuable.

Future Vision

HeHo will introduce a HeHo API key, allowing chatbots to be created directly from no-code and low-code apps. Future updates will enable AI to edit existing database entries based on permissions and even generate complete chatbot UIs using AI, enabling end-to-end AI products without writing code.

Note

Markdown In chatbots left

Built With

Share this project:

Updates