Customex

Company: Customex

Self-reported basis: The description is entirely from the author’s own submission to a hackathon — no independent verification, no archived history, no third-party corroboration.

What it appears to be: A browser extension and open protocol that allows users to request customised web experiences through AI, while respecting publisher-defined constraints.

What changed: The project was built as part of a hackathon; no prior version or commercial activity is evidenced.

Most important open question: Is there a viable path to adoption by website publishers who would be willing to publish manifests and support this kind of customisation?

The description states that Customex is:

• An open protocol.
• A browser extension (Chrome).
• A system where websites publish a machine-readable manifest.
• AI translates natural-language requests into reviewable, reversible view changes.
• The extension validates every action against the live manifest.
• Changes are explicit, atomic, and reversible.

The author describes it as a way to “bake customisability into a contract that is safe and easy to interface.”

Inference: It appears to be a developer-facing tool for enabling user-driven UI customization, mediated by AI but constrained by publisher-defined rules.

Not evidenced: No actual product demo, no customer feedback, no pricing, no technical architecture beyond high-level description.

The author states:

• The goal is to allow customised web experiences through AI, safely.
• It is built for web browsers, and uses AI (specifically Codex + GPT) to interpret user intent.
• It aims to prevent arbitrary code injection, while allowing publisher control over what changes are possible.

The project’s positioning appears to be:

• A protocol for safe, AI-driven UI customization.
• A browser extension that enables user interaction with websites in a constrained way.
• A developer tool (SDK) that allows publishers to define what customisations are allowed.

Inference: The author positions it as a safe alternative to full code injection, using AI to bridge intent and action.

Not evidenced: No market positioning, no competitor comparison, no evidence of prior user feedback or adoption.

The description states:

• The publisher (website owner) is the one who defines what changes are allowed via a manifest.
• The user makes natural-language requests, which are then translated into UI changes.
• The extension works for end users browsing websites that support Customex.

Inference: The primary customers are:

• Website publishers or developers who want to offer customisation options.
• End users who want to modify their browsing experience.

Not evidenced: No evidence of actual customers, no user personas, no publisher use cases, no market segmentation.

The description states:

• It is an open protocol.
• The author mentions a manifest-authoring assistant and a community library of safe view recipes.
• There is no mention of pricing, monetisation, or revenue streams.

Inference: The business model appears to be based on:

• Open-source or open protocol.
• Potential future monetisation through developer tools or services (e.g., manifest authoring assistant).
• Possibly community-driven adoption with optional premium features.

Not evidenced: No pricing, no monetisation strategy, no revenue model, no customer acquisition plan.

The description states:

• Built using Codex, GPT-5.6, JavaScript, and Chrome extension API.
• The manifest is published as a JSON element in the DOM to bypass content script isolation issues.
• AI translates user intent into patches, which are validated against the manifest before application.
• Changes are reversible, atomic, and require explicit approval.

Inference: The tech stack is browser-based with AI integration. It uses constrained AI to interpret user intent within publisher-defined boundaries.

Not evidenced: No architecture diagrams, no performance data, no scalability claims, no security audits or testing details.

The description states:

• This was built for a hackathon.
• The author is the only team member.
• It is an open protocol, but no evidence of adoption or usage by publishers or users.

Inference: The project is in early-stage development.

Not evidenced: No customers, no revenue, no user base, no product-market fit data, no traction metrics.

The description does not mention any competitors or similar products.

Not evidenced: No competitive analysis, no market landscape, no comparison to existing tools for UI customization or AI browser extensions.

• No commercial traction or adoption — it’s a hackathon project.
• No evidence of publisher interest — no real-world use cases or developer feedback.
• AI dependency — relies on Codex + GPT, which may not be scalable or reliable for production use.
• Limited team size — only one person built it.
• Unproven business model — no monetisation strategy or pricing structure.
• No technical validation — no performance, scalability, or security data.

1. What specific use cases have you identified for publishers to adopt this protocol?
2. How do you plan to incentivise website owners to publish manifests?
3. Have you tested the AI’s ability to interpret user intent reliably in real-world scenarios?
4. What are your plans for monetisation or commercial viability?
5. Are there any technical limitations or edge cases that could break the system?
6. How would you scale this beyond a single developer’s hackathon project?