§04 / PRODUCTS

Production-grade AI, ready on day one.

SIEVE and STRUCTURA put powerful AI in your team’s hands today: search that understands meaning, and agents that act on it for you. Deploy in the cloud or fully on your own infrastructure.

PRD/SIEVE

SIEVE

Semantic search & data understanding.

SIEVE searches and understands your data by meaning, not keywords, using our proprietary vector spaces. It indexes text, documents, and multimodal content into a semantic space where the right answer is the nearest neighbour.

SIEVE capability specifications
CAPABILITY DETAIL
Semantic search Query by meaning; nearest-neighbour retrieval
Multimodal indexing Text, documents, images, and more
Vector spaces Proprietary embeddings, tuned to your domain
Deployment Fully on-premise. Data never leaves your infrastructure
Talk to us about SIEVE
SIEVE on-premise architecture: data never leaves the customer perimeter A schematic. A dashed customer-perimeter boundary encloses the data store and SIEVE. A query vector snaps to its nearest neighbour inside the perimeter. An arrow attempting to leave the perimeter is blocked and labelled NO DATA EGRESS. CUSTOMER PERIMETER SIEVE QUERY DATA STORE NO DATA EGRESS blocked at perimeter
How it deploys: SIEVE runs entirely inside your infrastructure. Your data store and the semantic index sit within a customer perimeter; queries resolve to their nearest neighbour in-place. Nothing crosses the boundary, so there is no data egress.
SIEVE · SEMANTIC SEARCH RUNNING LOCALLY · NO DATA LEAVES YOUR BROWSER
SOURCES Documents Systems & databases Images Recordings Forward an email

Bring in data from anywhere: files, business systems, photos, audio. Or simply forward an email to your Causality inbox, and it’s ready to use in seconds.

SIEVE semantic search: a query resolves to its nearest neighbours, locally A 2D semantic space of twelve topic-clustered documents. A sample query is dropped in and its three nearest neighbours are highlighted with cosine similarity scores. All computation runs in the browser. SEMANTIC SPACE · nearest neighbour INFRASTRUCTURE SEARCH DATA PRIVACY MULTIMODAL 0.96 0.91 0.88
QUERY
  • on-prem
  • vector
  • database

RANKED RESULTS

  1. 01 on-prem vector database NEAREST 0.96
  2. 02 air-gapped deployment 0.91
  3. 03 kubernetes cluster config 0.88
  1. 01 / 05

    Feed it everything: documents, business systems, images, recordings. Forward an email and it’s searchable in seconds.

  2. 02 / 05

    Ask a question the way you’d say it out loud, with no keywords or special syntax.

  3. 03 / 05

    SIEVE reads it for meaning and places it on the same map as your data.

  4. 04 / 05

    The closest documents light up, even when they don’t share a single word.

  5. 05 / 05

    You get a ranked shortlist, most relevant first, and nothing ever leaves your systems.

PRD/STRUCTURA

STRUCTURA

Structured extraction, knowledge bases & agents.

STRUCTURA turns unstructured, multimodal data into a structured knowledge base, then runs custom agents that use that knowledge to take actions on the web and automate daily work.

Talk to us about STRUCTURA
STRUCTURA capability specifications
CAPABILITY DETAIL
Input Unstructured, multimodal data
Extraction Structured fields, entities, relations
Knowledge base Queryable, connected to SIEVE spaces
Agents / automation Custom agents act on the web, automate workflows
STRUCTURA · INTERACTIVE WALKTHROUGH

An example that runs on its own: emails arrive over time, the agent keeps your knowledge base up to date, waits for what’s missing, and acts on the web when everything lines up.

SOURCES Documents Systems & databases Images Recordings Forward an email

Bring in data from anywhere: files, business systems, photos, audio. Or simply forward an email to your Causality inbox, and it’s ready to use in seconds.

INBOX
Fwd: Order confirmation (Acme Foods)
PO.pdf pallet.jpg vm.m4a Supplier Order Quantity Amount Invoice
Watching, waiting for the matching invoice
Fwd: Invoice INV-2231 (Acme Foods)
STRUCTURA pipeline: from unstructured data to agents that act A left-to-right pipeline. Scattered document, image, and audio glyphs resolve into an ordered knowledge graph, which flows to an agent node that takes actions on the web. STRUCTURA supplies contains ships billed-by status Supplier Acme Foods Product Frozen Peas ×1200 Order PO-4471 Shipment SH-889 Invoice INV-2231 · €8,420 STATUS DUE AGENT watching acting on the web
AGENT ACTION
How it flows: emails and other sources arrive over time; STRUCTURA keeps a queryable knowledge base up to date, and a custom agent watches for what’s missing, reconciles it, and acts on the web to automate daily work. The scenario shown is an illustrative example.
  1. 01 / 05

    Data flows in from any source: files, business systems, photos, recordings, or a simply forwarded email. Everything lands in one place, ready to use.

  2. 02 / 05

    A first email arrives: Acme confirms order PO-4471. STRUCTURA reads it and records the order, product and shipment in your knowledge base.

  3. 03 / 05

    There’s nothing to pay yet, so the agent keeps watching, waiting for the matching invoice to arrive.

  4. 04 / 05

    A second email lands with invoice INV-2231. The agent matches it to the order, confirms the amounts line up, and marks it Due in the knowledge base.

  5. 05 / 05

    Now everything reconciles, so the agent opens the payment portal, enters invoice INV-2231 and schedules the payment, closing the loop end to end. This is an illustrative example, not a live transaction.