For a poset $P$, an ideal of $P$ is a downward closed upward filtered set of elements of $P$. The collection of ideals of $P$ ordered by inclusion is the free cocompletion of $P$ w.r.t. filtered colimits. If indeed the poset $P$ is a distributive lattice then one can simplify this definition a bit and define an ideal equivalently as a downward closed set which is closed under finite joins.
The appropriate generalization from posets to categories is that of ind-objects. An ind-object of $C$ is a diagram $D \to C$ where $D$ is a small filtered category. The free cocompletion of $C$ w.r.t. filtered colimits, denoted $Ind\text{-}C$ has ind-objects of $C$ as objects and for $F:D \to C$ and $G:E \to C$ two such ind-objects, the hom-set $Ind\text{-}C(F, G)$ is defined as $\underset{d\in D}{lim}\; \underset{e\in E}{colim}\; C(F d, G e)$ with limits and colimits computed in $Set$.
Now I need to know if there is any kind of simplification in case the category $C$ happens to be coherent. If we equivalently define an ind-object as a filtered colimit of representable presheaves there is a simplification: $Ind\text{-}C$ will be the full subcategory of PSh(C) consisting of those functors $F:C^{op} \to Set$ such that $F$ is left exact and satisfies a certain smallness requirement. So is there too a simplification for the definition of $Ind\text{-}C$ given above?
(My naive guess was to take as objects those diagrams $F:D \to C$ where $D$ has finite colimits and $F$ preserves them, but this obviously does not work.)
