Zenith

Inspiration

What it does

How we built it

Challenges we ran into

💡 Inspiration

In 2026, the primary consumers of your technical documentation are no longer just human developers—they are autonomous AI coding agents tracking down API specs, integration guides, and environment variables.

We realized a critical flaw: human developers can tolerate slightly outdated or poorly organized documentation. They can guess, check StackOverflow, or ask a colleague. AI agents cannot. If a parameter in your docs has drifted from your actual codebase, an agent will hallucinate, loop, and fail. We built Zenith because if your documentation isn't agent-ready, your product is practically invisible in the AI era.

⚙️ What it does

Zenith acts as a "brain" wrapped around third-party documentation services like Mintlify. It doesn't replace your documentation site; it supercharges it.

• Zero Maintenance (Drift Detection): Zenith continuously monitors your repository for codebase and API changes. When a modification happens, it automatically drafts the exact documentation update required.
• Agent-Optimized Metadata: It injects fresh, machine-readable metadata schemas directly into your files so LLMs can perfectly parse parameters, endpoints, and structural context.
• Aesthetic Preservation: We extract the underlying logic and structure from your .mdx files without touching your custom components, Tailwind classes, or animations. The docs stay beautiful for humans while becoming flawlessly structured for AI.

🛠️ How we built it

We architected Zenith as a middleware layer:

• Integration: We hook directly into Git providers and Mintlify repositories.
• Processing: We use AST (Abstract Syntax Tree) parsing natively for MDX to cleanly separate the "content/logic" layer from the "presentation" layer.
• Intelligence: We rely on structured LLM outputs to compare codebase diffs against current documentation, generating targeted patch updates.
• Delivery: Zenith automatically commits the enriched, metadata-injected, and updated Markdown back to the source repo, triggering the standard Mintlify deploy.

⚠️ Challenges we ran into

The hardest part was Aesthetic Preservation. Markdown and MDX files in modern docs are heavily intertwined with custom React components and styling. Stripping the text to rewrite it for AI often destroyed the visual layout. We had to build a meticulous parser that isolates the presentation layer, allows the LLM to rewrite the educational content and inject metadata, and then surgically stitches the visual components back together.

🏆 Accomplishments that we're proud of

We successfully achieved autonomous drift detection on a live repository. Seeing a deprecated API parameter in the codebase automatically trigger a contextual update in a corresponding Mintlify doc—while preserving all the complex CSS/animations on the page—felt like magic.

📚 What we learned

We learned that AI agents don't just need plaintext; they need contextual hierarchy. By injecting hidden metadata (like API lifecycles, exact endpoint schemas, and explicit versioning) directly alongside the human-readable text, agentic success rates on integration tasks skyrocketed.

🚀 What's next for Zenith

• Agent Analytics: Tracking which autonomous agents are reading your docs and where they are getting stuck.
• Expanded Integrations: Moving beyond Mintlify to support Docusaurus, GitBook, and ReadMe.
• Auto-Generated SDKs: Using the deeply structured documentation to automatically generate and maintain client libraries and open-source agent tools. ## Accomplishments that we're proud of

What we learned

What's next for Zenith

Built With

• gemini
• hermes-agent
• llm
• nextjs
• python
• typescript