Agentic Wijaywi

A WebMCP Provides a shared epistemic layer for AI agents to operate transparently on the open web, featuring hypothesis ledgers, automated falsification, and strict permission boundaries.

Comment

WebMCP-native human-agent collaboration environment. Agentic Wijaywi provides a shared epistemic layer for AI agents to operate transparently on the open web.

Deskripsi Lengkap (Untuk Formulir Submission Hackathon): Wijaywi Web Agent Reliability Lab adalah lingkungan kolaborasi native WebMCP yang dirancang untuk mengatasi masalah transparansi agen AI otonom. Berdasarkan arsitektur "Agentic Wijaywi", platform ini menyediakan antarmuka Epistemic Command Center di mana manusia dan agen berbagi world model, hipotesis, bukti, dan pembaruan keyakinan (belief updates) secara real-time. Melalui pendaftaran 10 WebMCP tools spesifik, aplikasi ini memastikan setiap tindakan agen berbasis bukti, dapat difalsifikasi, serta dibatasi oleh izin otorisasi yang ketat sebelum memanipulasi production state.

It's a WebMCP-native human-agent collaboration environment. It acts as an "Epistemic Command Center" where AI agents and humans share a world model. Instead of just chatting, the agents formulate hypotheses, gather evidence, update their confidence (beliefs), and explicitly ask for human approval before performing high-risk actions. All of this is powered by Next.js, SQLite, and native WebMCP tool registration.

Are you asking what to fill out in a specific form or hackathon submission field that asks "What are you building?" If so, you can use this text:

Agentic Wijaywi "A WebMCP native human agent collaboration environment that treats AI agents as persistent goal maintaining entities. It provides a shared epistemic layer for AI agents to operate transparently on the open web, featuring hypothesis ledgers, automated falsification, and strict permission boundaries."

========================================================================================

Inspiration

As AI agents become capable of operating web properties autonomously, we face a critical problem: lack of transparency. Agents often operate as black boxes, making it difficult to understand why they took a specific action, what evidence they relied on, or whether they hallucinated a conclusion. We were inspired to build a system that treats AI not as a generic chatbot, but as a goal-maintaining entity that must justify its actions through rigorous epistemic tracking.

What it does

Wijaywi Web Agent Reliability Lab is an "Epistemic Command Center" for human-agent collaboration on the open web. Built natively for WebMCP, it provides a shared environment where humans and AI agents can observe the same world state. Instead of blind execution, the system enforces a strict workflow:

  • Hypothesis Ledger: Agents must propose hypotheses before acting.
  • Evidence Tracking: Every claim requires timestamped evidence with clear provenance.
  • Falsifier Agent Role: Dedicated routines actively attempt to contradict existing beliefs to prevent confirmation bias.
  • Permission Boundaries: High-risk actions automatically halt and escalate to a human for explicit approval.

How we built it

We architected the platform using a modern, production-ready stack:

  • Frontend: Next.js 14 (App Router), TypeScript, and Tailwind CSS for the Mission Control interface.
  • State Management: Zustand for real-time UI synchronization.
  • Backend: Next.js API Routes connected to a SQLite database via Drizzle ORM.
  • Agent Integration: We utilized the WebMCP standard (document.modelContext.registerTool) to expose 10 highly specific tools directly to the browser environment.
  • Validation & Testing: Zod for strict schema validation and Vitest for unit testing.

Challenges we ran into

Integrating the WebMCP standard natively into a modern React ecosystem presented unique architectural challenges. We had to ensure that tool registration occurred safely on the client side while state mutations were securely synced to the SQLite backend. Designing a UI that completely avoids the "chatbot" paradigm and instead feels like a research laboratory required extensive iterations on data visualization and event streaming.

What we learned

We learned that making agent operations observable completely changes the dynamic of human-AI collaboration. When an agent explicitly lists its assumptions, current confidence level, and contradictory evidence, human operators can intervene with precision. We also discovered that WebMCP is a massive leap forward compared to traditional UI scraping, allowing agents to understand context and execute deterministic actions instantly.

What's next for Wijaywi Web Agent Reliability Lab

We plan to expand the uncertainty engine to support more complex causal relationships, introduce native PostgreSQL support for enterprise deployments, and build specialized UI modules for different agent personas (like the Synthesizer and Explorer).

Built With

Share this project:

Updates

Submission history