Vitrine AI : Recover and Finish Your AI-Built Projects

The description states that Vitrine AI is:

• A local-first project library and lifecycle manager.
• A tool that turns ChatGPT + GPT-5.6 into an evidence-cited project archaeologist.
• A system that uses a deterministic, read-only TypeScript scanner to discover project boundaries and surface contradictions.
• An MCP App exposing eleven narrow tools over synthetic judging projects.
• A system where GPT-5.6 inspects scan evidence, reconstructs project purpose, and submits structured assessments.
• A tool that supports isolated recovery: GPT-5.6 prepares a session-isolated copy, drafts a minimal fix, and shows a validated unified diff.
• Built with technologies including: chatgpt, codex, docker, electron, esbuild, gpt-5.6, mcp, node.js, react, three.js, typescript, webgl, zod.

Not evidenced:

• No actual product or service is described beyond the author's own development.
• No customer data, usage metrics, or real-world deployment details are provided.

The description states that Vitrine AI:

• Addresses a problem: “Finishing is not” after building with AI tools like Codex and ChatGPT.
• Positions itself as a tool where the archaeology is done by the model that helped create the mess — safely, with evidence, and without mutating source folders.
• Claims to be a local-first project library, emphasizing safety and privacy.
• Uses evidence-cited prompting to make AI output verifiable instead of merely convincing.
• Describes its workflow: scan → evidence → validated assessment → isolated diff → approved application → server-owned verification → library seal.

Inference:

• The positioning implies a niche in developer tooling for managing AI-generated projects, especially those with messy or inconsistent states.
• It is positioned as a solution to the problem of “AI archaeology” — resuming work on AI-built codebases that are hard to navigate.

Not evidenced:

• No prior versions or evolution history beyond this hackathon submission.
• No claims about market traction, user feedback, or competitive positioning beyond self-description.

The description states:

• The tool is built for developers who use AI tools like Codex and ChatGPT to build projects.
• It targets users who have “dozens of project folders: prototypes with three HTML variants, READMEs that contradicted the code, entrypoints nobody remembered, and half-finished ideas.”
• The intended audience includes those who want to recover and finish AI-built projects safely, without modifying source code.

Inference:

• Likely a subset of developers working in AI-assisted development environments.
• May appeal to users who are already using AI tools for rapid prototyping or experimentation.

Not evidenced:

• No specific customer segments, personas, or user interviews.
• No evidence of actual customers or target market size.

The description states:

• There is no OpenAI API call, no API key, no database, and no hardcoded assessment.
• ChatGPT is the reasoning host; Vitrine AI supplies verifiable evidence and validation.
• The author mentions a future roadmap including an OAuth 2.1 authenticated public /mcp deployment.

Inference:

• The tool appears to be built as a developer utility with no immediate monetization strategy.
• Future plans may include a hosted version or SaaS offering, but this is not yet evident.

Not evidenced:

• No pricing model, revenue streams, or monetization plans are described.
• No evidence of any paid features or commercial use cases beyond the author’s own development.

The description states:

• The tool uses a deterministic, read-only TypeScript scanner with explicit budgets (20,000 files, 32 MiB, 15 seconds per source).
• It leverages MCP Apps SDKs, Zod validation, and React + Three.js.
• Uses Electron desktop shell with a native folder picker and loopback-only WebGL Library.
• The system is built using Codex CLI with GPT-5.6.
• A 176-test suite ensures deterministic builds, widget integrity, and security hygiene.
• Recovery flow involves session-isolated copy, validated unified diff, and server-owned typecheck verification.

Inference:

• The tool is built for developers who value safety, reproducibility, and control over their codebase.
• It emphasizes privacy, isolation, and verifiability in its architecture.

Not evidenced:

• No evidence of production deployment or scalability beyond the author’s own use case.
• No details on performance benchmarks, latency, or infrastructure.

The description states:

• The project was built during a hackathon (OpenAI 2026).
• It includes a HACKATHON_BASELINE.md and HACKATHON_CHANGELOG.md to document provenance.
• A 176-test suite ensures deterministic builds and quality control.
• The author claims to have built the entire system in one week using Codex CLI.

Inference:

• This is a proof-of-concept or early-stage prototype, not a mature product.
• No evidence of user adoption, feedback, or real-world usage beyond the author’s own development.

Not evidenced:

• No customer data, usage metrics, or product-market fit indicators.
• No evidence of traction, growth, or market validation.

The description states:

• Vitrine AI is positioned as a tool for AI project archaeology.
• It uses ChatGPT + GPT-5.6 and MCP Apps SDKs to enable structured interaction with AI models.
• The system is designed to be local-first, read-only, and evidence-cited.

Inference:

• It competes in the space of developer tools for managing AI-generated codebases.
• It may overlap with tools that help manage or recover from messy development workflows, though no direct competitors are named.

Not evidenced:

• No competitive analysis or market positioning beyond self-description.
• No evidence of existing solutions or market gaps addressed.

The description states:

• The tool is local-first and does not use OpenAI API calls or databases.
• It uses evidence-cited prompting, which may limit the scope of AI reasoning.
• It enforces strict isolation to prevent source code mutation.

Red flags:

• No evidence of real-world usage or adoption beyond the author’s own development.
• The tool is described as a hackathon project, suggesting it is not yet production-ready.
• The use of GPT-5.6 and MCP Apps SDKs may be experimental or limited in scope.
• The lack of any monetization strategy or commercial plan raises questions about long-term viability.

Not evidenced:

• No evidence of user feedback, product-market fit, or scalability concerns.
• No indication of how the tool would scale beyond a single developer’s use case.

1. What is the actual problem you're solving, and how many developers are affected by it?
2. How does this differ from existing tools for managing codebases or AI-generated projects?
3. Have you tested Vitrine AI with other users beyond yourself?
4. What are your plans for monetization or product development beyond the hackathon version?
5. How do you plan to ensure that evidence-cited prompting scales and remains reliable in more complex use cases?
6. Are there any technical limitations or trade-offs in using a local-first approach?