Suppose a (continuous, non-negative) function $f$ satisfies $$ f'(t) \leq f(t)^2 $$ for $t \in [0,1]$. Set $$ g(t) = \exp\left(-\frac{1}{f(t)}\right). $$ Then $$ g'(t) = \frac{f'(t)}{f(t)^2} g(t) \leq g(t), $$ so applying the standard Gronwall inequality one has $g(t) \leq g(0) e^t$. Since $\log$ is an increasing function, applying $\log$ to both sides and rearranging gives $$ f(t) \leq \left( \frac{1}{1-tf(0)} \right) f(0). $$ Thus if $f(0) \leq \epsilon$ for $\epsilon >0$ small enough, we get $f(t) \leq C f(0)$ for all $t \in [0, 1]$. This proves a kind of small data nonlinear Gronwall inequality, but only starting from the differential form of the inequality.
My question: Suppose instead that we have $$ f(t) - f(0) \leq \int_0^t f(s)^2 \, ds. $$ Can we deduce the same result?
Here is an easy way to get the first part: $$\color{red}{f'(t) \leq f(t)^2} \Longleftrightarrow \frac{f'(t) }{ f(t)^2}\le 1 \Longleftrightarrow \left(-\frac{1}{f(t)}\right)'\le 1$$
integrating this latest gives
$$\left( -\frac{1}{f(t)}+\frac{1}{f(0)}\right)\le t\Longleftrightarrow f(t) \leq \left( \frac{1}{1-tf(0)} \right) f(0).$$