SYNTHEIA QUERY

Interrogate the clauses and definitions across your transaction corpus.

Most question-answering tools over legal documents give the model either everything or naive chunks of text, and hope the AI attends to the right clauses. That approach degrades as document sets grow, hits hard context-window limits, and gives you a probabilistic answer built on bad retrieval, rather than the actual clause.

Query is different. It navigates document structure — follows defined terms, resolves cross-references, tracks how amendments supersede base provisions — and returns the actual clause text, with the actual source, grounded in an explicit citation. The engine beneath every Syntheia instrument, made available to builders via API.

What the API returns
{
  clause_id: "4.1(b)(iii)",
  text: "IP indemnities — see Schedule 2",
  depth: 4,
  parent: "4.1(b)",
  source_doc: "Agreement_v3.pdf",
  page: 14,
  cross_refs: ["Schedule 2, §3.1"]
}

Not a summary. Not an inference. The actual text, at the actual provision level, with the actual source attached.

CONTEXT IS EVERYTHING

Searching and finding is the first step of solving the grounded context problem. Query is the next step.

Questions across a transaction set.

Credit facility agreements, limited partnership agreements, share purchase agreements — Query works across a corpus of transactional documents and lets an LLM find the answer wherever it lives, following cross-references and defined terms across documents without missing the connection.

Retrieve with logic, not just similarity.

Standard retrieval finds the clause that is semantically most similar to the query. It cannot represent the fact that multiple clauses should be retrieved. Query lets the LLM decide what to retrieve — returning the relevant provision, not just the highest-scoring neighbour in embedding space.

Defined terms and cross references.

A liability cap may invoke a defined term whose definition appears 50 pages earlier. An operative clause may point to a schedule containing the actual threshold. Query resolves both using inferred cross references — retrieving the clause, the definition, and the cross-referenced provision in a single pass.

The foundation for builders.

The same extraction and structuring engine that powers Compare, Curate, and our other instruments is available via API to legaltech companies and sophisticated firms and teams building their own products. Structured, hierarchical, source-attributed provision data — accessible via API or MCP.

What Syntheia Query gives you

  • Query navigates document structure rather than searching by keyword or vector similarity alone. Boolean semantic flags — does this clause contain a defined term? a date? — narrow the index before any similarity computation, the way a lawyer narrows a document before reading it.

  • Every node carries a crossReferencedIds list naming the provisions it references. The deterministic fetch follows these edges automatically — retrieving a clause also retrieves every cross-referenced provision and every defined term it invokes, without a second similarity search.

  • Full-corpus injection — placing every document in context on every query — degrades as corpora grow and hits hard context-window limits. Query retrieves only the relevant provisions, reducing token footprint by up to 30× and can match full-corpus accuracy on answer quality.

  • Every retrieved textblock is coupled with an explicit citation — source document, clause reference, or page number. There is no generative inference in the retrieval layer. Your model answers from the actual text, not its approximation of it.

SYNTHEIA Compare SYNTHEIA Curate SYNTHEIA Query SYNTHEIA Services