One database per region. Organizations, programs, service areas, events, jobs, places and records as rows with a review state, a source, and a public projection that only ever exposes reviewed public fields.
How it works
Community infrastructure with an intelligent layer.
LinkLocally is a locally owned community data platform: one place where a region can see what it already has. Organizations keep their own records current, a local partner reviews what publishes, and people find organizations, programs, events, jobs, seasonal guides and career paths by need and by place. AI reads, sorts and ranks; it never decides what becomes public. Northeast Wisconsin is the first region; the same platform repeats by configuration.
The work exists. The systems do not.
The homepage tells the story and shows the proof. This page is one level down: what happens inside each step, what the platform is made of, and what institutions can check.
The flow, one level down
Gather, check, publish, connect: what happens inside each step.
The homepage shows the four steps every record moves through. This is what each one is made of. Two of the steps are human, on purpose.
- GatherPublic records (the IRS filings behind 6,100+ organizations across 18 counties), partner information, public corrections from anyone, claims and updates from the organizations themselves, and, next, connected sources such as an organization's own calendar or careers page. Everything lands as a candidate, never as a published fact.
- CheckEvery candidate carries its source and a review state. Corrections and claims go to an operator queue; a local partner reviews its part of community life; a profile shows "Insight reviewed" only after that review and "Organization confirmed" only after a reviewed representative confirms it. AI helps organize the queue and flag likely changes. People decide.
- PublishWhat passes review becomes structured pages: profiles, category and county hubs, events, jobs, stories, the Summer Guide, Career Visibility. Records that have not passed the review their page requires stay hidden or clearly bounded. The same pages are emitted for search engines and AI assistants through sitemaps, structured data and llms.txt.
- Connect"Say what you need" reads a sentence and ranks the organizations most likely to fit, with reasons. Next: sources and dates on every answer, and a members' layer where the people behind organizations find each other.
Under the engine
What each of the six pieces is made of.
The homepage lists six pieces of the engine and marks each one live or next. Here is what sits under each label. The nonprofit directory is the first part of community life on the platform; the Greater Green Bay Summer Guide, Career Visibility and the Food Truck Finder are the same structure pointed at families, students and a moving local economy.
Three paths into a record: a no-account correction, a free claim managed with an email code, and Enhanced or Featured profiles with Insight on Philanthropy. One operator queue, an audit trail, and nothing that publishes itself.
Generated from the same records: the Insight on Philanthropy Nonprofit Directory with 31 cause hubs and 18 county hubs, events, jobs, stories, the Summer Guide in eight languages, Career Visibility, and the Food Truck Finder (prototype).
Roles in the database, not on paper: a partner stewards one part of community life with review rights scoped to it; the region owns its data; a new region is a new database with the same rules.
"Say what you need": the intelligent layer described below, reading one sentence and ranking organizations from their reviewed records.
Aggregate tables with a minimum-count floor, so partners and funders see coverage and gaps by county and by need, and nobody sees a person.
The intelligent layer
AI reads and ranks. People decide.
The intelligent layer does not write answers. It reads what a person typed, checks it against each organization's reviewed record, and hands a ranking to code that shows the reasons. There is no chatbot, and there is nothing for a model to invent. Its first job is a person's need; the same layer is built to read whatever sentence a community page accepts, from a student's to an employer's.
1 sentencein, a ranked list with reasons out, under two seconds when warm
0 wordsgenerated; every result is a record a person can check
Under 1¢per question, so the public layer can stay free
Nothingtyped is stored, logged, or used to train anything
AI is the electricity: wired into every step, replaced by a better source every few years, and never the reason anyone walks in. What stays is the structure, the stewardship, and the connections made.
Next for the layer: results grouped by need so an answer reads as a path, hours and eligibility on each service, transit proximity from the public feeds the region's bus agencies already publish, and job and childcare openings synced from the organizations' own sites. Each is an addition to the data, not a new product. Try it now.
Built like infrastructure
What institutions can check today, and what is next.
Partners, funders and public agencies ask the same questions of any technology they adopt. Here are the answers as they stand, so nobody has to guess. The eight trust rules that govern every page are on the homepage; the review states are defined in the verification policy.
In place today
- Security at the database level: row-level access rules by region and role, verified from the anonymous side; operator, partner and organization roles; email-code sign-in.
- Governed operations: claim, correction and review workflows with an audit trail; nothing publishes itself; every AI-touched capability behind a fail-closed switch.
- Backups with a restore rehearsal that passed; automated code-security scanning, secret scanning, an accessibility gate and a performance gate on every release.
- Bounded AI: the intelligent layer stores nothing typed, generates no text, and runs under per-address and daily limits.
- Repeatable by configuration: one database per region, the region owns its data, partner-branded deployment with the same rules.
- Open to machines: structured entity pages, llms.txt, sitemaps, and a published verification policy.
Not yet, and we say so
- Single sign-on for institutions.
- A written service-level agreement.
- Third-party security certification.
- Formal procurement documentation.
- A public API and an MCP server with provenance on every field.
- A second region in production.
Who owns what
Owned locally. Stewarded by people who know the place.
Questions about the platform, a partnership, or a deployment in another region: hello@linklocally.org.