How iri handles your knowledge
iri organizes information in three steps. It's more than "wiki with good search" — each step transforms the material so people and AI can use it the same way.
Step 1. Storage (Episodic)
What goes in. PDFs, meeting notes, Slack threads, Notion pages, your own notes — original material as-is.
What iri does. Stores the original and indexes it as smaller chunks for search. Search runs two ways:
- Keyword search — matches on the exact words.
- Semantic search — also picks up similar phrasing. ("quarterly revenue" finds "Q2 revenue".)
When you look at this directly. When you need the original document or a citation. Most of the time the upper layers give you a finished answer.
Step 2. Distillation (Semantic)
What goes in. Single facts pulled from the originals. iri calls these atoms.
For a sentence like "Q3 launch date is set for September 15," iri produces:
- About: "Q3 launch date"
- Value: "September 15"
- Source: "2025-08-22 decision meeting.pdf"
Why this matters. As your company grows, decisions on the same topic end up scattered. Storing them as atoms means iri can pull them together to answer one question.
More. Atoms
Step 3. Answer (Working memory / Briefing)
What it does. When a person or AI asks a question, iri assembles the relevant material into one organized response. We call this a briefing.
Example query:
"What did we decide about security for project X?"
iri returns:
- Current decision — OAuth + passkeys (April 22 meeting)
- Related material — five documents: decision notes, policy spec, security review
- Related context — adjacent security discussions from the same period
Not a list of search hits — a freshly assembled answer for the specific question.
More. Briefings
At a glance
| Step | What lives here | Who looks |
|---|---|---|
| Storage | Original documents | When you need the source or a citation |
| Distillation | Single facts pulled from documents | iri uses these to build answers |
| Answer | An assembled response per question | People searching, AI tools calling in |
The three steps together mean you don't have to remember "where to look." iri assembles the answer every time.