Changelog

Every map that counts things now uses one consistent style of sized dots and numbered clusters, and the cities and countries listings can switch between a grid and a map.

Places without a full profile now show a tidy facts page, places without a cover get a drawn street plan in your site's style, and you can set a place's cover photo from its admin page.

The Reuse atlas grew from a few dozen city radars to 149, adding cities across Spain, France, Italy, the Netherlands, Belgium, Germany and Central Europe, each drawn from that city's own official registers.

Contradictions found on several German city profiles were fixed, figures now always match the source they cite, and broken links on Irish and Scottish profiles were re-sourced.

The conversion check now compares a building with confirmed works costs from built conversion projects, adjusted to today's prices and set against new build in each country — and every radar city is wired to it.

Fixes that kept production deployments from failing, fewer paid preview builds, and leaner automated checks.

Year of construction now comes from building registers, crowd-sourced maps and, as a last resort, an estimate from street photos — clearly labelled as such.

Local news reports of empty buildings are now gathered for all 134 radar cities, cleaned of duplicates and placed on the map; a new gallery shows lighthouse conversions that already worked.

Numbers about the Reuse atlas inside deep dives — buildings on record, cities covered, lighthouse projects — now come straight from the atlas and refresh as it grows.

A new pilot campaign page invites building owners, partners, companies, investors and founding members, and every route leads to a single role-driven sign-up form.

The weekly atlas run now publishes newly found buildings to the live site on its own, behind a safety check; any building can be found by name or address in the radar search.

When a member updates their profile on the site, the change is sent to your Notion workspace, and AI-written bios are checked against the member's own data.

Stakeholder groups are now the one way to group people and organisations, members choose their own groups, and each group can be asked its own set of questions.

A large clean-up removed unused code, tightened automatic checks and fixed small display bugs such as lists losing their bullets.

Building cards gain a floor area, storey count, year built, listed status, daylight and — for public bodies, companies and churches — who owns it. Private owners are never named.

Each admin and write action now checks the signed-in user itself, sign-in can no longer be faked with a cookie, and old public routes that could be misused are gone.

The participation form now changes an existing account only when the person proves they own it — by being signed in or following their own invitation link.

From five German cities to over fifty across Europe — Paris, Vienna, Milan, Barcelona, Copenhagen, London and many more — each built from that city's own open registers.

A conversion check on every building answers first — does it fit, what could go in it, and what could a co-op afford — using the economic model's own numbers.

The economic model lives on one page, with a fair-shares engine, the real mid-stay rental market, a reworked cost breakdown and funding schemes matched to the right region.

The new Reuse atlas maps empty and under-used buildings building by building, with a shortlist of where to look first, a guided 3D tour and shareable links to any building.

All 100 city profiles are live in their newest version, Germany's 30 largest cities now have profiles with local rents, and the sources they cite feed the resource library.

Error and "page not found" pages, the cookie banner and Notion-based pages were reworked, and the links to your imprint and privacy policy now come from your site settings.

A cover inbox gathers AI-checked image suggestions for your resources in one place, and new resources reliably get a cover.

Newsletters, digests and automated emails are drawn in the same visual identity as your public pages.

Visitors can reach you through a built-in contact form, and embedded YouTube videos load only when someone presses play.

Filtering, searching and sorting the resource library now happen on the server, so the page stays fast as your library grows.

Every scheduled job records each run, and a watchdog flags any job that never ran in your inbox and in a daily alert.

Public pages are served from cache again, ship only the icons they use, and no longer download the whole library to show a list or a map.

Every proposed change is now checked automatically, quiet failures are logged instead of ignored, and new tests cover admin access, library filtering and the cost model.

Around 100,000 lines of unused scripts, unused web addresses and dependencies were removed, and duplicated helpers for reading Notion, caching, image search and AI models were merged into one of each.

The map, sitemaps, chatbot, navigation and admin pages now fetch only the library details they need, resource pages are prepared on the server, and each page loads only the icons it shows.

Visitors can now write to you through a contact page that forwards straight to your contact email without storing anything. Newsletter sign-up is no longer pre-ticked, and the site stores far less personal data.

Each admin and data-changing action now verifies your sign-in itself, risky unused public addresses were removed, and a check stops new unprotected ones from being added.

Every nightly and hourly job now records each run in the same way, and a watchdog flags any job that has missed its last two scheduled runs — on your admin Home and in a daily alert.

Every email the platform sends — newsletters, digests, drip sequences and notifications — now draws its look from one shared style, and the email branding editor lets you pick a design and preview it.

The daily documentation updater no longer publishes invented edits, and its changes are kept rather than silently overwritten.

The admin menu is now grouped by what you are doing — Library, Editorial, Places, Community, Communications, Site, AI and Operations — and opens on a Home inbox listing what needs your attention. Press Cmd/Ctrl+K to jump to any admin page by name.

The admin area is regrouped into sections named for what you are doing — Library, Editorial, Places, Community and so on — with a Home inbox, role-based access and a command palette.

Research runs now search the web on every step instead of answering from memory, finished results land in a "Ready to publish" inbox, and the "Advance" step works again with a watchdog for runs that stall.

The economic model now reads as one story — the building, the organisation, then growth — with a "Go deeper" option in each part, a public list of every assumption, and a way for readers to say they disagree.

The city profiles now cover Germany's 30 largest cities, with measured rents and local funding schemes, and dead citations in held drafts were repaired before publishing.

Research runs that had stalled at the Advance step move again, and a watchdog now catches runs that go silent.

Both the text chatbot and the voice assistant now draw on all of your authored pages, not just the library, and the voice panel keeps a notebook of the conversation.

The platform now fetches only the data each page needs, rebuilds cached pages less often and skips needless builds. Tenants' databases and analytics share one server, cutting monthly running costs.

The public site now has a 'cut & paste' visual identity: paper textures, tape, a hand-drawn logo, matching fonts and one consistent set of buttons, tabs and cards. It is switched on per tenant through the theme, so other communities keep exactly the look they had.

The cities section now covers Europe's capitals and largest cities with researched, sourced profiles, each on its newest version. Facts were audited city by city and every profile now reaches your Notion.

The chat and voice assistant has been redesigned to match the new look, shows a live caption paced to its voice, and cites where each answer came from.

AI now labels who each knowledge source is for, so the Knowledge overview can hide very local sources and show sources in the visitor's languages.

The vision page and the 'how it works' story can now be read or watched: a calm AI voice narrates while the drawings animate in step, and each word is highlighted as it is spoken. Listeners can pick 0.75× to 1.5× speed.

Voice conversations keep one session at a time, captions keep pace with the voice and the language no longer flips on a stray word.

Drafts are now saved before their links are checked, the link check has a time limit, and it no longer mistakes foreign-language or bot-protected pages for broken ones. The library also stops filling up with duplicate resources.

A new Illustration Studio generates illustrations in your site's style, alongside cuts in database traffic.

Every published deep dive can now be played as audio read by an AI voice, split into chapters. As it plays, the current word is highlighted and the page scrolls along with the narrator.

Every deep dive can now be played as audio by an AI voice, with the text highlighted as it is read.

Every deletion of a stored image or file now goes through one guarded check that refuses anything still in use, and a nightly check alerts the team if a cover ever goes missing.

Smaller fixes from the quarter: correct "date added" sorting, projects keeping their location, hero videos playing in more browsers and steadier builds.

Cities like Köln and Kraków now open from any spelling, and the city picker and duplicate check recognise all of a place's names.

City and country addresses are now built consistently from their names, so links with accents reach the right page instead of a 'not found' error, and any of a city's names finds it in search and pickers.

Deep dives now come out of a reviewable ideas pipeline, and each one gets a generated hero image when it is published.

An optional chaptered homepage walks visitors through why housing is hard in their city, how a cooperative network answers it, who it is for and how to join. You choose between it and the Notion-based homepage from one setting.

Filtering, searching and sorting the library now happen on the server instead of in the visitor’s browser, cutting what each visit downloads by roughly 35 times and removing a fault that could leave the page showing a month-old copy of your library.

When a profile cites a report or study, that source is added to your library as a proper resource rather than left as a bare external link, so a cited paper becomes something your members can find, search and browse.

The research agent was moved onto Perplexity’s current interface ahead of the 27 September shutdown of the one it used before, so scheduled research runs carry on uninterrupted.

Research, topics, the economic model and two new reference pages on funding and leasehold now sit together in a single Explore and Understand section, with long-form deep dives that link straight through to the evidence behind every claim.

City and country profile runs no longer hang open after finishing, the AI cost of each run is now reported accurately, and a batch of contradictory figures across 24 city profiles has been corrected.

The model now separates membership contributions from the shared fund, applies relief for households joining in later cohorts, and shows a clearer household-level figure.

All web research moved to Perplexity's new Agent service before the old one was switched off on 27 September.

The everyday chatbot tier and the voice assistant now run on Gemini 3.8, giving better answers at the same cost, with the model choice still adjustable from your admin settings.

A site-wide setting had been quietly stopping every public page from being cached since late April; with it removed, city, country, resource, news and topic pages are served from the cache nearest the visitor.

The Economic Model gained a switch that routes surplus from established buildings into reducing what a later group of households pays to join. It reports the trade-off rather than hiding it: a lower one-off buy-in, but a slightly higher monthly cost.

An audit found three assumptions shown on the model's research page describing something different from what the model actually calculates, so they were corrected and the panel carrying them was retired. The automated checks guarding the published figures went from 20 to 60.

The database is no longer held awake by constant polling, log tables are pruned automatically, and background jobs can now be paused for one tenant without touching the others.

Each part of a city or country profile — timeline, quotes, linked resources, section text — now refreshes on its own schedule instead of the whole profile being rewritten at once, and approved profiles are mirrored back into your Notion Cities and Countries databases.

The weekly assistant that writes these entries is now handed a single-use credential that works only while its run is in flight, instead of holding a long-lived shared secret.

Guides that have fallen behind the platform are detected, rewritten from how things actually work now, and put up for a human to approve — around a dozen were refreshed this way over the past month.

The weekly changelog writer now receives a single-use credential that expires with the run, rather than relying on a long-lived secret stored in the configuration.

The research paper behind the Economic Model is now assembled straight from Notion, so edits there flow through to the published version and the frozen snapshot people cite. All 36 figures were re-typeset to stay readable in print, and five had their underlying data or sources corrected.

Everyday AI work — search, chat, summaries — now goes to a cheaper, faster model, with only genuinely hard questions escalated to an expensive one, cutting the monthly AI bill by roughly two thirds. The database has stopped being held awake around the clock, old log entries are cleared on a schedule, and each community can now switch off the background jobs it does not use.

The pipeline that turns production errors into fixes now reports each result the moment it finishes instead of waiting for the next hourly check, and a new admin panel lets you decide how confident it has to be before it fixes something without you. Errors that only ever happen on a developer's machine are filtered out.

Errors thrown by browser extensions and other third-party scripts no longer turn into bug reports, and a completed fix now updates the pipeline instantly instead of waiting for the next hourly check. You can also tune how incoming bugs are sorted from a new panel on the auto-fix page.

The automatic bug-fixing pipeline now hears back the instant a fix is finished instead of waiting for the next hourly check, and it no longer keeps retrying work that has already been abandoned.

An old file-based copy of the data, a quantity of dead code and a set of needlessly interactive page sections were all removed, cutting what a visitor's browser has to download and clearing a hard release-size limit.

Uploaded PDFs and linked papers now have their text pulled in for search and their first page rendered as a preview image, and a free-to-read copy of a paywalled paper is found automatically where one exists.

The Research Agent now chooses its AI model on measured results rather than assumption, ranks candidates correctly, and its review queue handles a batch of decisions in a fraction of the clicks.

Uploaded PDFs now have their complete text captured rather than just the opening pages, get a cover picture rendered from their own first page, and where a paywalled paper has a freely readable version the platform finds and links it. Suggested cover images are validated before they reach you.

The research agent now picks its AI model on the strength of measured results rather than assumption, a ranking fault that buried good candidates is fixed, and the review screen was rebuilt so you can work through a batch of suggestions in far fewer clicks.

An old layer that kept a duplicate copy of your content in files alongside the database has been removed entirely, along with a large amount of unused code. Pages now read from one source of truth and send less to the visitor's browser.

A voice conversation now resumes after a drop, always shows its wrap-up summary at the end and settles its budget correctly. Separately, the AI models behind the chatbot were rebalanced to cut the monthly bill by roughly two thirds.

A wide reliability and speed pass on your voice chatbot: a typical spoken exchange now returns in about a third of the time, sessions survive a network blip instead of ending, wrap-up summaries always render, idle open-mic tabs stop billing your allowance, and members get their unused voice minutes back after a failed start.

A dedicated /how-it-works landing page that tells the Economic Model story as a scroll-through visual narrative — pinned scenes, hand-coded illustrations, a properly projected map of Europe, and a two-year building conversion that the reader unveils at their own scrolling pace.

A performance and cost pass on the research agent: repeated or overlapping searches now share a two-tier cache so you stop paying twice for the same discovery work, the admin runs list loads faster, and stalled items surface a clear reason instead of sitting quietly stuck.

A new admin dashboard meters every AI call — chatbot, research, image checks, geo lookups, transcripts, admin tools and voice — and rolls up cost per day, per feature and per tenant.

A live, city-by-city economic simulator covering financing, ownership pathways, rent build-up and pilot funnel — 155 European cities, three market scenarios and a citeable frozen thesis preset, now fronted by a guided story for first-time visitors.

City and country profiles got a v2 layout, a richer data catalogue (Loop 38 + Themes 1-4 → 13 new fields, 496 new cells), an in-Notion 'Refresh from AI' button and a two-way related-content service that keeps cross-links in sync.

Resolved a multi-day production deploy outage by moving to enhanced 16 GB build machines and tidying client/edge bundles. Deploys ship in under four minutes again, reliably.

A quarter of small but visible improvements: rich previews when you share links on social media, smoother mobile filters on cities and countries, and dozens of small fixes across admin and public pages.

City and country profiles now include curated rent levels, currency conversion for cost comparisons, and a coverage view that shows where data is still thin.

Each night, the platform pulls curated city and country data from a shared reference dataset and mirrors the fields you've subscribed to into your tenant's Notion.

Every topic on the platform has its own page that gathers featured resources and opens with a short AI-written introduction grounded in your community's own content.

Instead of waiting for community members to submit, you can ask the platform's Research Agent to find events, organisations, projects and research papers on a topic — anywhere in Europe.

Knowledge resources, organisations and projects now get an AI-written summary backed by real, clickable citations to the original sources.

Every mention of European Housing Coop was replaced with per-tenant configuration, so the same codebase can host different communities with their own branding, content and feature mix.

You can publish news articles from Notion with proper bylines linked to your community members, set up through a guided wizard for new tenants.

A weekly AI routine audits the platform documentation against recent code changes, reads the actual code, updates the docs and opens a pull request to ship them.

Your community can now monetise premium features — Stripe handles checkout, individual features can be locked to subscribers, and revenue analytics show subscription growth and churn.

Server and browser errors are now captured automatically, deduplicated, and synced into a Notion database where admins can triage and track resolution.

A consistent set of admin building blocks (cards, buttons, toggles, tabs, badges, tips) now powers every admin page, replacing the one-off styles.

The homepage moved from hardcoded markup to a structure where every section is edited from Notion and synced to the live site within seconds.

Nightly automated backups run at 3:30 AM UTC, plus a major housekeeping pass to clean up legacy debug files.

Four parallel browser test sessions, each auto-authenticated, now let the team — and AI agents — exercise admin features end-to-end without manual setup.

A complete newsletter system with a visual editor, signup widget, automatic Notion sync, drip campaigns and unsubscribe handling.

Set up email through your provider of choice — Amazon SES, Postmark, Brevo or Loops — and edit every template in a drag-and-drop visual editor.

Adding a YouTube video now reliably captures its transcript and high-resolution thumbnail, with multiple fallback strategies when YouTube's own transcripts are missing or malformed.

A complete SEO stack: a sitemap built from the live database, structured data on every resource and location page, RSS feeds, and an admin dashboard showing which pages need attention.

The chatbot grew from a basic widget into a full conversational assistant with voice input/output, persona variants and admin analytics.

Events published on the platform now appear on a Google Calendar that anyone can subscribe to from iCal, Outlook or Apple Calendar — direct sync, no third-party middleman.

A quarter of stability work: automatic recovery from development cache corruption, first-load rendering fixes, email editor refinements and keyboard navigation across the admin.

Submit a Knowledge resource (PDF or URL) and the platform extracts the full text, summary, title and cover image into Notion automatically.

Every AI surface — chatbot, enrichment, image verification, research — upgraded to the current Gemini generation, with a monthly check that surfaces newer models when they appear.

Persistent caching combined with pre-rendered pages at the edge bring city, country, resource and topic pages from 2–7 seconds down to 400ms–1s.

One admin screen to see every member, change roles, run bulk actions and re-sync from Notion People — plus a profile completion flow for members and a custom invitation system.

Every colour on the platform — primary, secondary, accent, success, warning, error — is configurable from one admin page. Dark mode included, contrast guaranteed.

The admin area was reorganised around a clear sidebar grouped by purpose (Content, Users, Communications, Design & Layout, Settings, Analytics, System), with a search palette and breadcrumbs.

Every scheduled job, sync run and webhook now reports to a central dashboard with success/failure counts, durations and metadata.

The Notion integration was upgraded through three API versions over the year, keeping the platform on Notion's current feature set without breaking any tenant.

The Image Discovery tool searches Google, Wikimedia, Unsplash and Pexels for any resource, then verifies each candidate with AI vision before showing it to you.

Events, organisations, projects, knowledge and media all got continuous polish over the year: faster pages, sharper cards, better image handling and smarter search.

When an admin edits a resource, city or country in Notion, webhooks now trigger an immediate sync so the public site updates within seconds — no waiting for the nightly run.

Major performance improvements, real cover imagery, proper filter UX, share controls, and tight linking between resources, events and organisations in each location.

A single polished form adapts to whichever type the contributor is submitting (event, organisation, project, knowledge, media), and detail pages got proper share controls and better cover imagery.

The /participate form gained full field coverage, country and relation handling, redesigned UI, session-mixing fixes and graceful handling of empty databases.

The database evolved through the year — new tables for embeddings, knowledge content, audit logs, drip campaigns, subscriptions, geo replicas — all backed by safety scripts that warn before destructive changes.

Three months of refinements: image manager improvements, PDF title extraction, event filtering, layout polish and dozens of bug fixes across every surface.

A semantic search system powers the chatbot's ability to answer natural-language questions about your platform's actual content.

A pass through the platform's rendering layer cut page payload sizes, fixed runtime issues on the edge, and stopped the rendering glitches that occasionally caused authentication problems.

A complete sign-in system with Google and LinkedIn OAuth, email-based magic links, proper session management and multi-tenant isolation.

The Notion sync gained exponential retry (up to 5 attempts), rate limiting, deletion safety checks and timestamp-based change detection.

The early Notion-sync GitHub Actions workflows were retired and replaced by scheduled jobs running directly inside the platform.

The platform's first months in production: hundreds of small fixes, deployment hardening, Notion ID resolution and image caching.