SceneBoard

SceneBoard, as described by its author, is a visual workspace that enables Codex (a large language model) to explain complex software development work through structured, interactive, and presentation-grade artifacts. It operates as a plugin and Model Context Protocol server, connecting to Codex via a one-time pairing mechanism with explicit capability grants.

The product aims to expand context for decision-making by rendering internal references like D1, D2, or ADR-014 into accessible visual formats—such as diagrams, charts, HTML prototypes, WebGL experiences, and interactive decision workspaces. It supports Human-in-the-Loop (HITL) workflows where human decisions shape subsequent AI outputs, and all revisions are preserved immutably.

SceneBoard is built as an Apache-2.0 TypeScript monorepo, using technologies including Next.js, NestJS, React, Three.js, and GPT-5.6. The system includes sandboxed artifact execution, a capability-based security model, and a UI that supports both visual and textual content.

Key commercial due-diligence read: The author states SceneBoard is a working end-to-end product at sceneboard.dev, but there is no evidence of revenue, customers, or adoption beyond the project's own description. The core innovation lies in how it allows Codex to operate visually while preserving HITL decision control—this positioning may be novel, but its commercial viability and traction remain unproven.

The description states that SceneBoard is:

• A visual whiteboard designed for Codex to explain complex software development work.
• A plugin and Model Context Protocol server, enabling Codex to create and operate boards directly from CLI or IDE.
• An Apache-2.0 TypeScript monorepo with four main layers:
  • Next.js web application
  • NestJS API
  • SceneBoard Codex plugin and MCP server
  • Shared protocol schemas, SDKs, rendering components, and artifact runtime

It allows Codex to generate various types of visual content including:

• Structured documents, tables, diagrams, architecture flows
• Charts, SVGs, animated Canvas scenes
• Interactive HTML prototypes
• WebGL and Three.js experiences
• Human-in-the-Loop decision workspaces

The system uses a sandboxed artifact origin with a constrained capability bridge to ensure security.

Inference: SceneBoard appears to be a visual collaboration layer for AI agents, particularly those operating in development environments, aimed at improving transparency and decision-making during AI-assisted coding.

The author positions SceneBoard as:

• A second surface for Codex, expanding context beyond terminal output or source files.
• A visual workspace where Codex explains work clearly and humans make informed decisions.
• A tool that turns dense AI work into presentation-ready briefings, supporting HITL workflows.

Key claims:

• “SceneBoard is the whiteboard Codex can operate freely.”
• “Codex builds. SceneBoard explains. Humans decide.”
• “A useful board must support interactive HTML and 3D content, but that content must never gain silent access to accounts or credentials.”

These statements suggest a shift from AI summarization to contextual explanation and visual collaboration, with an emphasis on security boundaries and human agency.

Inference: SceneBoard positions itself as a visual extension of AI development workflows, not just a tool for summarizing code, but one that supports decision-making through visual artifacts.

The description does not explicitly name target customers or define an Ideal Customer Profile (ICP). However, it implies:

• Users who work with AI agents like Codex in development environments
• Developers, product teams, engineers, operators, and decision-makers involved in AI-assisted coding workflows
• Teams that need to understand and review AI-generated code or decisions, especially when internal references (e.g., D1, ADR-014) are involved

It also suggests a developer-first approach—the system is built with open-source direction, contributor guidelines, and engineering practices.

Inference: SceneBoard likely targets technical teams using AI agents in software development, particularly those seeking better context and decision-making support during AI-assisted work.

There is no evidence of a business model or pricing structure in the description. The project is presented as a self-contained open-source tool, built by one person (Dexter Lee), and submitted to a hackathon.

The author mentions:

• An Apache-2.0 license
• Public engineering, security, Git, and QA rules for contributors and AI agents
• A monorepo with public documentation

Inference: SceneBoard is currently not monetized, nor does it appear to have any pricing or commercialization strategy described.

The description provides technical details:

• Built on TypeScript, Next.js, NestJS, React, Three.js, WebGL, GPT-5.6
• Uses Model Context Protocol (MCP) and plugin architecture
• Implements a sandboxed artifact runtime with capability-based access control
• Supports expressive visual output: HTML, Canvas, SVG, WebGL, 3D content
• Includes pairing flow with one-time approval, using short-lived codes
• Has immutable revision history for boards and decisions

It also notes:

• Self-hosted development workflow where Codex helped build the product itself
• Use of regression tests around artifact rendering, decision correlation, and pairing recovery

Inference: SceneBoard shows strong engineering maturity for a hackathon-level project. It demonstrates security-conscious design, modular architecture, and visual expressiveness—all critical signals for a developer tool.

The description states:

• A working end-to-end product at sceneboard.dev
• The system is open-source under Apache-2.0
• It was built in a single-person effort, with Codex contributing to its own development
• The author mentions regression testing, browser-tested desktop experience, and multilingual UI

However, there is no evidence of users, customers, revenue, or adoption metrics beyond the project’s own submission.

Inference: SceneBoard has reached a functional prototype stage, but lacks any traction signals such as user base, usage data, or commercial interest.

The description does not mention competitors. However, based on its stated functionality:

• It operates in the space of AI collaboration tools
• It integrates with AI agents like Codex
• It supports visual explanation and HITL workflows

Potential areas of overlap include:

• Tools that visualize AI outputs (e.g., Notion, Obsidian, GitHub Copilot)
• AI agent interfaces or workspaces
• Human-in-the-loop systems for AI decision-making

Inference: SceneBoard appears to be a novel approach within AI development tools, but there is no evidence of existing competitive landscape or market positioning.

Key risks and red flags based on the description:

• No revenue, customers, or traction—the project is self-reported and unverified.
• Single-person team—limits scalability and long-term maintenance.
• Unproven commercial viability—no pricing, monetization strategy, or business model.
• High technical complexity—requires sandboxing, capability control, and multi-layered architecture; potential for security or performance issues.
• Limited evidence of real-world use cases—the author describes a vision, but no actual deployment or feedback from users.

Inference: SceneBoard is technically ambitious, but lacks commercial validation and real-world traction. Its success depends heavily on adoption by AI developers and teams using Codex.

1. What specific use cases have you identified for SceneBoard in real-world development workflows?
2. How do you plan to scale beyond a single developer or team?
3. Have you tested SceneBoard with actual users or teams? If so, what feedback did you get?
4. What are the key challenges in integrating SceneBoard into existing AI agent workflows?
5. Is there any intention to monetize or commercialize this tool?
6. How do you plan to handle security and access control at scale?
7. What is your roadmap for expanding visual artifact types or HITL decision models?