Your Docs Are Lying to Your AI Assistant. This Dubai-Based Startup Wants to Fix That.
Every developer has hit the same wall: an AI coding assistant confidently suggests a method that was deprecated two years ago, or references an API endpoint that no longer exists. The code moved on. The documentation didn’t. Thally, a startup bootstrapped from Dubai by developer experience leader Ekene Eze, has now publicly launched to close that gap, turning product changes directly into pull requests for the documentation, websites and support content that describe them.
A problem that got worse, not better, with AI
It’s tempting to assume AI tools would make documentation drift less of a problem. Thally’s founder argues the opposite is true. Stack Overflow’s 2025 Developer Survey found that 84 percent of developers now use or plan to use AI coding tools, yet only 29 percent trust their accuracy, with hallucinated APIs and long-deprecated methods among the most common complaints.
The reason traces back to content that quietly falls out of date and keeps circulating anyway. An outdated integration guide or code sample stays publicly discoverable long after a product has moved past it, and increasingly, it isn’t people finding it first. Mintlify’s July 2026 traffic report put 66 percent of documentation traffic at 213 million requests a month, from AI agents rather than human readers, against 105 million human page loads over the same period. A stale page doesn’t just confuse a developer skimming for an answer; it misleads every coding assistant that crawls it, at scale.
Eze knows the problem firsthand. “At several companies, the teams I led owned the documentation, and I did this work myself: hours spent reconstructing what a change meant, then hunting for every public page it touched,” he says. “Now that product changes ship faster, and AI tools read the docs more than people do, that gap is more expensive. Thally does the tracing and the drafting, and leaves the judgment and the final approval with the team.”
How it actually works
At the center of the product is a feature called Thally Track. Teams connect their product repositories to the content repositories they want kept in sync, whether that’s documentation in Markdown or MDX, or websites built in React. Thally can run on a set schedule or trigger directly off pull requests, either when one opens or when it merges.
When a relevant product change lands, Thally works through four steps: it compares the exact before-and-after commits to identify which connected content the change may affect, closing out anything outside the paths a team has chosen to watch; it drafts the update inside a disposable, isolated sandbox that holds no repository, database or deploy credentials and is destroyed once the run ends; it validates that draft by running documentation checks and a full site build before proposing anything; and it opens a pull request listing exactly what changed, which product commits it compared, which source files it reviewed, and the results of its checks.
Crucially, Thally never merges or deploys its own pull requests. The team still reviews the proposal in GitHub, edits it as needed, and decides what actually gets published. That review layer is deliberate; Thally is positioning itself as the tool that does the tracing and the first draft, not the one making the call.
More than a bot that opens pull requests
Alongside Track, Thally ships a full AI-native documentation platform, built on an MIT-licensed engine that teams can self-host if they choose. A single MDX source generates every format a reader or an AI agent might need: rendered pages, per-page Markdown, JSON, JSON-LD, an llms.txt file, and a Model Context Protocol (MCP) server that tools like Claude, Cursor and Copilot can query directly.
The platform also includes managed hosting with versioned deployments, pull request previews and instant rollbacks; hybrid full-text and semantic search that works with zero configuration; AI-generated answers with citations verified against the retrieved source before they’re shown to a user; analytics that separate human visitors from AI agents; an Agent Readiness Score that grades a documentation site from 0 to 100 and names the specific pages dragging the score down; localization across 34 languages; 42 built-in themes; and one-click framework upgrades that don’t overwrite a team’s existing content or configuration. Teams migrating from Mintlify, Docusaurus, GitBook, Nextra, VitePress or Starlight can bring their existing repository or live site straight in.
From developer relations to founder
Eze began building Thally in March 2026, after more than seven years in developer experience and developer relations roles at companies including Flutterwave, Netlify and LI.FI. He’s running the company solo and fully bootstrapping it from Dubai, a detail that places Thally in an increasingly visible cluster of African and Africa-connected founders building infrastructure-layer developer tools from outside the traditional Silicon Valley pipeline.
Since releasing its beta in August 2026, Thally has grown to more than 100 active workspaces, counting teams currently on its 14-day trial, with beta users spread across the United States, Europe, India, the Middle East and Africa. Design partners on the product include Hackmamba.io, Procta.org and LI.FI.
William Imoh, CEO of Hackmamba, put the appeal in blunter terms: “We write and maintain technical content for developer-focused companies, and the hardest part has never been the writing. It is knowing which pages a product change just made wrong,” he says. “Thally watches the repositories and shows up with a pull request that already lists the commits and files it based the update on. Our team reviews instead of reconstructing, and the docs stay trustworthy for the developers and the AI tools reading them.”
Getting started
Thally is live now at thally.io, with a free tier that supports one documentation site, including managed hosting, previews and analytics. Thally Cloud costs $199/month or $1,990/year and bundles Thally Track, AI answers, custom domains, three managed sites, five team members, and 10,000 AI credits monthly. Enterprise plans add SSO and SAML, SCIM provisioning and audit logs. Every new workspace can start a 14-day trial of Thally Cloud without entering a credit card.
For a problem that’s easy to describe and expensive to ignore, that’s a fairly low bar to test whether Thally actually closes the gap between what a product does and what its documentation says it does.
About Thally
Thally is the product knowledge layer for software teams. It connects product repositories to documentation, websites and other customer-facing content, traces the impact of product changes across those surfaces, and opens evidence-backed pull requests for human review. The platform also includes a full AI-native documentation platform with custom domains, versioned deployments, instant rollbacks, AI answers and one-click site upgrades. The documentation engine behind Thally is MIT-licensed and can be self-hosted. Learn more at thally.io. For media inquiries, contact founder Ekene Eze at [email protected].
