Saving something to AiAkiv creates three things. Knowing who makes each one tells you what actually deserves attention.
| What it is | Who makes it | How much it matters | |
|---|---|---|---|
| Event | An account of something (one piece of what happened) | The user's input | ▽ It's simply what was saved |
| Entity | A key thing appearing in it | AI-extracted + the user's | ● Important: search lives here |
| Tag | An extra handle for finding it | AI-suggested + the user's | ● Important: the axis you slice by |
Event: the account the user saved
One save is one Event. Save "we moved combat to turn-based" and that's an Event. The Event is the text the user wrote. On save, AiAkiv attaches a short summary and Entities to it automatically. Search hands you the summary; when you need more, you pull the full text.
What matters is that an Event doesn't sit alone. Entities and Tags weave it to the others. That weaving is the whole of AiAkiv.
Entity: why it matters
An Entity is a thing that appears in the event: a person, a feature, a file,
a concept. In "we moved combat to turn-based", combat system and turn-based
are Entities. They're extracted automatically on save (NER).
Why it matters: Entities are the knots that tie events together.
- If five events share the Entity "combat system", those five form one thread. → "How did the combat system change?" is answered by that whole thread.
- If an inventory event and a save-file event share the Entity
serialization schema, then "why did those two change together?" is answered through that shared knot.
One rule to follow: one name per thing. Always call the same thing by the same name. Mixing "save format" and "savefile format" splits one Entity into two and cuts the knot. Keep names consistent and the graph weaves itself. (→ Saving)
Tag: why it matters
A Tag is a label the user attaches. If an Entity is "who and what appeared",
a Tag is "what kind of event this is": decision, bug, balance, retro.
Why it matters: it becomes the axis you slice by later.
- "Show me just the decisions" / "just the bug history". You need Tags to cut that way.
- Attach both a kind (decision / bug…) and a topic (combat / save…) and you can narrow on the intersection.
The AI suggests a few Tags on save, and the user can add to them or fix them. Whether the suggestions are left alone or tuned, that small bit of involvement comes back as search quality.
At a glance
- Event is the account the user saved. The AI adds the summary and Entities.
- Entity is the knot that ties events together (AI-extracted, plus the user's). Save with one name per thing and search comes alive.
- Tag is the axis. The AI suggests, the user tunes. It pays off when you slice later.