Meaningful word IDs may improve agent memory and cut token use
Long numeric IDs were replaced with related to each item's . Instead of following common naming rules, the system now uses readable names similar to wiki page titles. This was practical because all data stays local and is handled only by the agent and its .
The nodes and edges in the agent's were also renamed from numbers to . During one week of use, the agent appeared to stop responding that it could not find stored information, and its memory seemed better. Token use also seemed to fall sharply because the repeatedly places many IDs into the agent's input.
However, this is a short l experiment without measured results or a controlled comparison, and the method may not suit every system.
Key points
- Replace long numeric IDs with short tied to the .
- Use the same readable names for nodes and edges in a .
- The agent appeared to have no failed lookups during the one-week test.
- Token use seemed to drop because the supplies many IDs.
- No exact measurements or controlled test results were provided.