Skip to main content

Documentation

Documentation is how AI Expedite turns the work you ship into customer-facing material — release notes, help articles, API references, and changelog posts — and keeps that material in sync as the code underneath it changes.

📷 Screenshot: A repo's AI-generated documentation block, expanded inside the Codebase tab.

Unlike the other pillars in Launch, Documentation isn't a single in-app tab today; it's the output of several systems that work together. This page is the orientation map.

Where docs are generated​

Three surfaces produce documentation across the platform:

  • Per-repo AI documentation — every repo in Code → Codebase has an expandable AI-generated overview: what the repo does, key modules, primary domain, integration points. The platform refreshes this on a cadence by re-analyzing the codebase, and the Re-analyze action on a repo accordion forces an immediate refresh after a big change.
  • ChangeLog summaries — every commit that lands on a default branch gets an AI-written change summary in Code → ChangeLog, classified by type (feature / bugfix / enhancement / …). These summaries are the raw material for release notes.
  • Workspace documents — anything finalized for human or customer audience lands in General → Documents as a typed doc (Requirements / Design / Meeting / Other). Agents draft into these, owners approve, and the resulting Tiptap rich text is the canonical version.

What gets produced​

The documentation agents combine those three sources to draft, in order of automation level:

  • Release notes — from ChangeLog entries scoped to a release window, grouped by classification (features first, fixes next, internal/chore last). These post automatically to the channels you've connected in Communication and to any docs site you publish out.
  • Help articles — from feature specs in the originating Roadmap card plus the per-repo AI docs of the code that implements the feature. Drafted into a workspace document of type Other, owner-reviewed, then promoted.
  • API references — from the per-repo AI documentation plus the actual route definitions / schemas the agent reads from the codebase. Re-runs whenever the underlying code changes, so the published reference doesn't drift.
  • Internal changelogs — quick "what shipped this week" digests that the Communication agents post into team channels.

Each of these is produced by a system-agent hook you can rebind in Agents → System Agents if you want to customize voice, length, or which sources get used.

Keeping docs in sync​

The "kept in sync with the code that backs them" part comes from two mechanisms working together:

  • Re-analyze on commit — when a commit lands, the platform queues a re-analysis of the affected repo's AI docs and any docs that reference its symbols. Stale references surface in the originating workspace document with a "needs update" indicator.
  • Living documents — the Tiptap editor in General → Documents supports inline diffs from agents, so when an API rename happens, the relevant help article gets a proposed edit rather than silently going stale.

Documents that have gone untouched since their backing code last shipped get flagged in Inbox under the Document category, so you can spot the drift before a customer does.

Publishing out​

The platform doesn't host a docs site for you — the canonical pattern (and what docs.aiexpedite.com itself uses) is:

  1. Connect a docs-site repository in Code → Codebase.
  2. Tell the documentation agent to write into that repo's content directory.
  3. Use whatever static-site generator you prefer (Docusaurus, MkDocs, Mintlify, Astro Starlight, …) — the agent matches the format the repo already uses.
  4. Standard CI on that repo deploys the site.

Because the docs site is just another repo in your workspace, every other Ship surface (Tests, Deployments, Monitoring) applies to it too — broken links fail tests, deploys promote new content, and uptime checks watch the live site.