iri docs

Atom — a single fact

An atom is one fact pulled out of a document. Your meeting notes, policies, and decision logs are hard to search as-is. iri stores the originals and also distills the key facts inside them. Each distilled fact is an atom.

Example

Suppose a meeting note has this line:

"Redis runs with a 4GB-per-instance limit."

iri stores this as an atom:

Field Value
About Redis
Relationship operating limit
Value 4GB per instance
Type policy
Source 2025-09-12 infra decision meeting.pdf

Why split it like this. Later, when someone asks "what's our Redis usage limit?", iri can answer with the exact line and source instead of dumping the whole meeting note.

Atom types

iri tags each atom by what kind of fact it is:

  • Fact — objective, doesn't change
  • Decision — who decided what, when
  • Ownership — who's responsible
  • Status — current state of progress
  • Dependency — what depends on what
  • Policy — rules and guidelines
  • Definition — meaning of a term or concept

These tags let you scope a search ("only show me decisions").

What happens over time

Company information changes. iri tracks atoms this way:

  • Reinforced atoms get more confidence. When a fact is cited often or new docs repeat the same content, its confidence climbs.
  • Unreferenced atoms fade. Older, unreferenced material drops in priority over time.
  • Pinned atoms freeze. Things that shouldn't fade can be pinned by a person, exempt from time-based decay.

The auto-confidence and contradiction features are still being refined. For high-stakes answers, have a person verify.

How this differs from plain search

Most search tools return document chunks — a paragraph that mentions "Redis."

iri goes one step further: a single fact extracted from a sentence. So a question like "Redis operating limit" gets the value (4GB) and the source in one shot.

Document search + fact distillation = the answer quality keeps improving as iri sees more material.

See also