How Stratta works
Why Stratta gives you exact, verifiable citations instead of plausible-sounding guesses.
Stratta is designed around one goal: every answer must be traceable to the source norm. For engineering work, an answer you can't verify is worse than no answer. This page explains the approach in plain terms.
The problem with ordinary AI search#
Most AI document tools chop a document into thousands of small fragments, turn each into a numeric vector ("embedding"), and retrieve the fragments that look mathematically similar to your question. This works for casual Q&A, but it has two failure modes that matter for norms:
- Fuzzy provenance: A retrieved fragment may not map cleanly to a section number or page, so the citation is approximate or invented.
- Lost structure: Norms are deeply hierarchical and cross-referenced. Flat fragments lose the chapter → section → clause structure that tells you where a rule actually lives.
Stratta's approach: TreeRAG#
Instead of embeddings, Stratta stores each norm as its real table of contents: a tree of chapters, sections, sub-sections, and annexes. Claude navigates that tree the way an engineer flips through the printed norm:
See what's available
Claude lists the norms in your workspace.
Open the table of contents
It reads the chapter structure, then drills into the relevant section.
Read the exact section
It opens the section's full content: text, formulas (in LaTeX), tables, and figures.
Follow the cross-references
SIA splits a rule across norms: the action in SIA 261, the safety factors in SIA 260, the material rules in the domain norm. Claude follows those links before concluding.
Because each section keeps its exact path and page range, every citation Claude produces
([SIA 261, 8.2, p. 44]) points to a real, openable location. This is what we mean by
deterministic citations.
TreeRAG is inspired by the PageIndex approach to document reasoning. Read the deeper rationale in Concepts → TreeRAG.
The two halves of Stratta#
Where you sign up, create API keys, manage your organization, and (later) handle billing.
The MCP server you install in Claude. It authenticates with your API key and gives Claude the tools to read and ingest norms.
There is no separate chat interface to learn. You use the Claude client you already have; Stratta simply adds the norms.
Your norms are yours#
Norms are scoped to your organization. You ingest the documents you are licensed for, and they are visible only to your workspace, never a shared public corpus. This keeps you in control of licensing and confidentiality. See Concepts → Multi-tenant norms.