Guides
Guides and case studies
Real projects and task-shaped recipes. Each one says what we did, why, and where WREN still falls short.
Start here
Take an org live
Slug, keys, data model, public rules, first release, your domain, AI agents, edge settings and backups, in order.
ConceptsDocuments, versions, labels and trees
The mental model: immutable versions, labels as pointers, how a URL picks a version, and why promote is atomic.
TutorialDeploy a site in ten minutes
Deploy, preview with a label, promote to published.
How-to recipes
Publish a site and keep drafts private
Public rule with a label filter, deploy as preview, promote atomically, roll back.
How-toFind your org slug and build public URLs
Private path to public URL, which form to share, and what 401, 403, 404 and 405 mean.
How-toUpdate data and files without duplicates
Upsert by natural key, replace files in place, and clean up orphans.
PatternsI did this. Is there a better way?
Workarounds real apps used, from date twins and search-then-create to preview copies and shared owner keys, each with the better way today.
How-toUse WREN from Python
Public and private reads, upserts, files, cursor pagination, and the 403 "error code 1010".
How-toConnect AI agents with MCP
The built-in /mcp endpoint: set up Claude, Cursor or VS Code, and let agents write to preview while humans promote.
How-toServe a site on your own domain
A small Cloudflare Worker that maps your domain onto a public tree, without an API key.
Case studies
A live tournament tracker
Live taekwondo brackets and results from Python pollers and a static site, as built (warts included) and rebuilt on natural keys, materialized queries and atomic promote.
Case studyAn open tournament data site
Entry lists and ranking seeds as open JSON from a scheduled pipeline, rebuilt on natural keys, unchanged writes, ISO date filters, promote and org-bound MCP.
Case studyA webshop and its warehouse stock tracker
Barcode stock counts in private collections, copied to a shop's database by hand, rebuilt on webhooks, narrowed keys, files by name and collection restore.
Case studyA task tracker that AI agents write to
Projects, tasks and a work log written mostly by coding agents, and how MCP, per-agent keys, live events and deep diffs fit it.
Case studyWebsites maintained by AI agents
Four sites deployed and checked by agents: staging by copy and overwrite, rebuilt on preview labels, atomic promote, restore to a release and files by name.
Case studyRepeatable test data with versions
An API emulator whose test data is labeled and restored after every run, from a loop of hundreds of requests to one transactional restore.