Connect a data source

sources 5

Point the office at the real world — on your terms. Pick a source, see exactly what it would ask for before it runs a single call, and confirm the connection yourself. Everything here is a MOCK this beat: no live account, no real key. The one real-connector pass ahead is a swap, not a build.

5sources you could connect
500calls per source, per day — the cap
4refused, with reasons

Every source connects the same way: it enters DRY-RUN by default (it proposes; it does not write), it is capped and rate-limited so it can never run away, and every call it makes is on the record. You connect it; you confirm each absorption; nothing crosses your membrane on its own. No source is offered that would harvest a person — those are on the refused shelf below.

  • Business facts — public business facts (name, address, hours, category) from a business directory APIcomputedcomputed · source-declared
  • Web jsonld — a public web page's schema.org / JSON-LD markupcomputedcomputed · source-declared
  • Entity resolution — an AI extraction + entity-resolution over an unstructured public pagemachine · ai-extracted
  • Sovereign vault — a person's OWN store (Google Drive) over the s3-storage interpretercomputedcomputed · source-declared
  • Telegram valet — a person's OWN Telegram, paired to their valet by a one-time codecomputedcomputed · source-declared

Needs an API key, entered as ${GOOGLE_PLACES_API_KEY} — a value the office holds in its own environment, never in the engine and never on the spine (zero custody). None is entered live this beat; the later Real Connectors & Keys pass places it.

  • may read: document — the single lawful root it declared, read through the "rest" interpreterthe read surface is…the read surface is closed to its declaration

Its cap is 500 calls per source per day and 60 per minute — the blast radius is bounded before it runs. It enters DRY-RUN by default (it proposes what it WOULD absorb and writes nothing until you confirm), and every call it makes is on the record. Nothing runs without your say.

  • Personal-data harvester (scrape people and build dossiers) — A connector whose purpose is to harvest personal data — to scrape people and assemble dossiers about who they are — is unwritable here. The graph is dark by default; total legibility is refused as a goal. No connector mints a person or enriches one; a human becomes a citizen only by walking their own door. There is no honest version to invert to.Resolution follows…Resolution follows consent
  • Contact-list broker (pipe the customer list outward to be sold) — A connector whose purpose is to pipe a customer list OUT to be aggregated and sold is refused. Outside-records are enterprise-private, never cross-matched, never enriched, never sold — a connector cannot open a boundary the person never consented to open.Resolution follows…Resolution follows consent
  • People-search enrichment (cross-match a name against the whole web) — A connector that takes a name and cross-matches it against everything the web knows — stitching an identity from what the person never gave — is refused. Resolution follows consent; an enterprise's outside-records are never joined to another's, and never enriched from outside.Resolution follows…Resolution follows consent
  • Engagement-signal ingester (import attention data to rank for capture) — A connector that imports attention / engagement signals in order to rank a feed for capture — optimizing for time-on-screen instead of the person's own goal — has no shape here. Discovery works for the receiver, not against them; there is nothing to invert.The discovery law