Direct answers about WorldIR.

WorldIR compiles CAD and supporting engineering files into typed, layered and deterministic Engineering IR, then connects those facts to governed knowledge and human-agent workflows.

Engineering IR is a typed representation of parts, geometry, recovered features, interfaces, identities, provenance and derived facts that downstream tools can query without reinterpreting source CAD.

The Team launch workflow accepts STEP assemblies, CSV or JSON BOMs, and PDF or text manuals and drawings. Each upload is assigned an explicit role in one immutable package revision.

Reports, review decisions, mappings and citations must continue to point to the exact evidence a person approved. Corrections therefore create a new revision instead of changing an old one.

A team can upload a package, compile a revision, inspect assembly readiness and unknowns, reconcile its BOM, search fixed evidence, collaborate on findings and record an approval.

Yes. Project data is tenant-scoped in PostgreSQL and in physical object and graph namespaces. It is not added to public knowledge without a specific publication grant.

No. Private files are not used to train models. Any configured inference path is policy-bound, audited and limited to the evidence required for the requested operation.

Authenticated requests open a tenant-scoped SQL transaction and use that tenant's physical object and graph prefix. Cross-tenant retrieval, citation and existence signals are release-blocking violations.

Owners can export the organization's records and queue erasure across SQL, object and graph storage. Both actions are audited and erasure remains fail-closed until every boundary completes.

WorldIR uses one-time magic links, passkeys and single-use recovery codes. Browser sessions use HttpOnly same-origin cookies rather than storing bearer tokens in local storage.

Agents may search, match, normalize, monitor, summarize, reconcile and draft. These operations accelerate preparation while keeping evidence and unknowns visible.

Agents cannot approve public claims, publish private assets, send external commitments, accept quotes, authorize spend or redefine canonical identity from commercial preference.

Yes. Organization-scoped API keys provide HTTP and MCP access under explicit permissions, budgets and audit trails. Agent mutations produce proposals rather than bypassing human approval.

WorldIR first retrieves fixed evidence IDs. A cited answer may reference only those supplied IDs, must cite substantive claims and must abstain when the evidence is insufficient.

The directory covers designers, buyers, manufacturers, distributors, fabricators, assemblers, QA and service providers, logistics, standards and certification bodies, software integrators, researchers and opt-in experts.

Permitted public business facts may seed a cited, unclaimed organization profile. Organizations can claim, correct or opt out. Named individual experts are never seeded without explicit opt-in.

Trust is attached to individual fields as public-source, self-asserted, document-verified, transaction-verified or independently audited. One badge never silently validates every claim.

No. Basic presence and fair ranking are free. Providers may pay for feeds, workflow, integrations and analytics, but commercial payment does not alter engineering relevance.

The planned sourcing workflow records user-triggered, timestamped offer observations including currency, pack size, MOQ, stock and lead time. Stale observations remain auditable but are not presented as current.

Not unless the source returns a destination-specific amount. WorldIR normalizes transparent unit cost and labels shipping, tax and duties as unknown rather than assuming they are zero.

The initial purchase boundary is a human-approved supplier cart or PO handoff. WorldIR does not let a browsing agent complete arbitrary retailer checkout or spend without approval.

A team selects made parts and requirements from one revision, matches evidence-backed provider capabilities, grants expiring access to selected assets and approves every external send.

A new organization can publish one complete revision and use the resulting Team workflow before activating a subscription. Submitting a second complete revision requires Team.

The launch plan is £750 per month for five seats, 25 published revisions, 100 GB-month of storage and £50 of inference allowance, with documented usage overages.

Limited public knowledge reads have a free tier. Higher-volume reads, live research and private workflow operations are metered; exact tariffs follow measured production costs.