Decagon builds AI agents that resolve customer support conversations for major enterprises.
Decagon made Mintlify its source of truth and built a self-healing system on top that now keeps its documentation, internal knowledge, product slides, and an open industry specification current as the product changes.
95%
of documentation updates now driven by automation
200%
increase in PRs merged each month
10+
teams use the internal docs agent to manage knowledge updates
Challenge
Keeping docs current was a manual process with no clear owner, and internal knowledge was often outdated and scattered across tools. Decagon's in-product AI agent, Duet, answers customer questions from the company's documentation, so anything out of date could reach users directly.
Solution
Decagon made Mintlify a source of truth for both its product and internal company knowledge, then built an agent on top that drafts updates as the product changes. The docs stay current, so Duet answers more customer questions and internal teams get reliable answers about best practices, processes, and how deployments actually work. That same source of truth now extends past the docs, powering the slides teams present from and the open protocol specification Decagon publishes for the industry.
Multiple sources of truth
Decagon understands the value of a good knowledge base. But when David Hou joined to lead the company's education and enablement initiatives, keeping Decagon's product knowledge current was still a manual, ad hoc process with no clear owner.
The gaps showed up in the product. Because Decagon's in-platform agent, Duet, answers customer questions from its own knowledge base, every missing or outdated page directly impacts the customer experience.
A similar issue was happening internally. Best practices, processes, and learnings from real customer deployments were scattered across internal sources. Anyone asking Claude or Codex a question risked getting a stale or incomplete answer.
I knew we needed a single source of truth for all company knowledge. Not only is this critical for ensuring accurate, up-to-date answers for customers and employees, but it also serves as a foundation for us to build agentic workflows, tools, and experiences on top.
– David Hou
Product Education Lead at Decagon
A system that heals itself
Because Mintlify is a docs-as-code platform, Decagon's documentation lives as markdown in a GitHub repo that agents can read and edit like any other codebase. That allowed Decagon to build an internal agent of its own.
The agent watches for triggers across the company, including commits landing in the codebase, product announcements in Slack, and anything else signaling that knowledge has changed, and takes a first pass at updating the content. Custom skills and an AGENTS.md file enforce Decagon's writing guidelines. Engineers, PMs, and Hou crowdsource the review, and a human publishes every change.
Mintlify makes it extremely easy to connect to other tools and build custom automations. Because everything is maintained as code, even the docs.json file, I've been able to build a system where documentation and knowledge as a whole can self-heal and self-update as the product changes.
– David Hou
Product Education Lead at Decagon
Today, an estimated 95% of Decagon's docs updates are automated. Hou also estimates that the agent saves employees 10-15 hours per week, all while 200% more PRs are merged each month.
Knowledge as a product asset
With a single source of truth that stays current on its own, Decagon can treat knowledge as an integral part of its platform, rather than a byproduct. Since Duet's launch, its knowledge coverage has greatly increased and customers are feeling the improvements.
"Agents are the new audience for your knowledge, and that comes with real product requirements: how it's shaped, what it contains, and how retrievable it is," shares Hou. "When we improve knowledge, Duet gets better at its job, and users of our platform feel that directly."
Hou has since productized the docs agent inside Decagon. What started as a single documentation automation is now a full-fledged agent that maintains both product and internal documentation, including a second Mintlify deployment holding internal best practices, processes, and deployment learnings. Other teams with their own knowledge to maintain can point the same system at it, with 10+ teams across the company having adopted the tool so far.
"Internal-facing knowledge is equally as important as external-facing," notes Hou. "Using Mintlify as a source of truth for our knowledge base helps teams find the right information faster and more confidently, so they can ultimately do their jobs and serve our customers better."
From docs to infrastructure
The same source of truth that powers Decagon's docs also powers decks used by teams from across the company, including GTM, Product, and Deployment. Product slides regenerate as the product changes, so sales and deployment teams work from the same current information as the documentation.
This is what knowledge looks like as operational infrastructure: the same system that answers a customer's question in the product also arms the team selling and deploying it. The material that closes deals and educates end users is only as good as what a company actually knows about itself, and at Decagon that's now maintained automatically rather than assembled by hand every quarter.
Because Mintlify keeps everything as code, our source of truth isn't trapped in a documentation tool. We can build any form factor on top of it: docs, internal knowledge, now slides. It stops being documentation and starts being infrastructure the whole company runs on.
– David Hou
Product Education Lead at Decagon
A spec the industry builds on
The latest form factor documents a standard rather than a product. In October 2026, Decagon and Instinct open-sourced the Personal Agent Consent & Trust Protocol (PACT), a specification for how personal AI agents act on a customer's behalf with businesses. It lives at openpactprotocol.org, built and hosted on Mintlify, with every page available as markdown and indexed in an llms.txt file.
Agents that learn from what they know
Decagon set out to uplevel its documentation and ended up building something the whole company uses. The docs maintain themselves, Duet answers more of what customers ask, and teams across the business use the same system to keep their own knowledge current. Now, an open industry specification benefits from it too.
The form factors are just getting started. Slides today, demos next, and after that the thing Hou is really building toward: agents that improve from the same source of truth they answer from.
The most AI-enabled companies only exist because they have great knowledge. Good knowledge means good products. And I believe good knowledge starts with Mintlify.
– David Hou
Product Education Lead at Decagon