Per-message model stamping
Short answer: Every assistant reply is saved with the exact model and provider that produced it, and a small tag under the bubble shows it. A reply with no recorded model shows no tag rather than guessing.
What is model stamping?
A chat isn’t locked to one model for its whole life: you can switch models
mid-conversation, and each reply is generated by whatever model was active
at that moment. Build Perch records that fact permanently: when a reply is
saved, its model and provider are written into the message itself, not
inferred later from the chat’s current setting. If you switch a chat from
one model to another, older replies still show the model that actually
wrote them.
How do I see which model wrote a reply?
- Send a message and read the reply as usual; no action needed.
- Look under the reply bubble: a small
model · providerline names exactly what generated it (hover it for the full names in a tooltip). - If you switch models partway through a conversation and scroll back through the history, a labeled divider marks the point where the conversation handed off from one model to the next, so you can see the whole chat’s model history at a glance.
Good to know
- If a reply has no recorded model, Build Perch shows no tag rather than labeling it with the chat’s current model, which could be wrong. It never guesses.
- The stamp records the model that actually ran. If OpenRouter’s router (e.g. an “auto” model) routes your request to a different underlying model than requested, that’s what gets recorded.
- Media replies are stamped too. Generated images, audio, and video carry the same idea (the model that produced them), shown alongside the result.
- There’s nothing to configure; stamping is always on and reflects reality automatically as you switch models.