What does 'confidence bounds for the predicted values' means in context with 'logit'-link function?

I am using 'glmfit' and 'glmval' -functions with 'logit' -link function to perform logistic regression analysis, having the response values 0 or 1.

My understanding is that the output 'yhat' from 'glmval' with 'logit' link function stands for probability in range [0:1].

In manual 'yhat' is said to present 'the predicted values'. Then, how must be the 'confidence bounds for the predicted values' 'dyhi' and 'dylo' be interpreted? They can not be summed directly on the 'yhat', since that would make the probability <0 or >1 in some cases.

In this case, how should I calculate the confidence bounds for the probability 'yhat'? I am using Matlab r2007b.

Edit: My guess is, that the confidence bounds relate to the output of the link-function, while the output of the 'glmval'-is processed further to give the probability value.

Set the output of 'glmval' = yhat = p. Then boundary values relate to the boundary probability as: ln(p/(1-p))+dyhi = ln(phi/(1-phi)), and ln(p/(1-p))-dylo = ln(plo/(1-plo)).

From these, the probability boundaries can be calculated, and the result seems to look as expected. If this is the case, then Matlab presents yhat and (dyhi, dylo) in different domains; yhat is probability, and (dyhi, dylo) relate to ln(odds).