(Drift table)
How agents behave on each model.
When a team moves an agent to another model, Kyde sees what changes in its behavior. Here is that view across the Kyde fleet, never one customer's agent.
- Models in the table
- 0
- Gathering data
- 4
- Window
- 90 days
- Data as of
- 21 Nov 2026, 15:00 UTC
Snapshot from 21 Nov 2026, 15:00 UTC. Loading live data…
| No model has enough data yet. 4 models are gathering data. A model appears once agents from at least 5 different Kyde keys run on it. | ||||
How agents behave per model, aggregated over the Kyde fleet: switches of the last 90 days, each observed for 7 calendar days after the switch; a row appears only when at least 5 distinct keys contribute, and never shows a single customer's agent. This measures agent behavior on a model, not model quality.
- Agents on this model
- Agents in the Kyde fleet that ran on the model in the last 90 days.
- Switches
- How often an agent moved to this model in that window. Each switch is watched for 7 calendar days.
- Flagged within 7 days
- Of those switches, how many ended with the agent in Watch or Deviating within 7 days. These are Kyde Score bands: the agent behaves noticeably differently from its own normal.
- Median score change
- The middle value of how much the agents' Kyde Score moved in the 7 days after a switch, in points from 0 to 100. Below zero means agents behaved less like themselves.
- What it does not say
- It does not rate the model. It shows how agents that already run in the Kyde fleet behave on it. A model with few switches tells you little either way.
- Privacy
- A row appears only when at least 5 different Kyde keys contribute. Test keys never count, and the fleet numbers never change an individual agent's score.
(Your agent)
Switching models? Watch your own agent before and after.
The table shows agents in general. Kyde Score learns how your agent normally behaves and tells you when it changes after a switch.