The free groupoid on a directed graph is the groupoid whose objects are the vertices of the graph and whose morphisms are finite concatenations of the edges in the graph and formal inverses to them.

Given a graph$D$, that is, a collection of vertices and of labeled arrows between them, the free groupoid $G(D)$ on $D$ is the groupoid that has the vertices of $D$ as objects, and whose morphisms are constructed recursively by formal composition (i.e., juxtaposition) from identity maps, the arrows of $D$ and formal inverses for the arrows of $D$.

The only relations between morphisms of $G(D)$ are the necessary ones defining the identity of each object, the inverse of each arrow in $D$ and the associativity of composition. This is clearly a groupoid, which comes with an evident morphism $D \to G(D)$ of quivers.

The above sketched construction could be made more precise, but what really matters is the universal property it enjoys: the free groupoid $G(D)$ is the universal (initial) groupoid mapping out of $D$. By varying $D$, the free groupoid yields a functor$G$ from directed graphs to groupoids, left adjoint to the forgetful functor.

This last conceptual characterization is best taken as the definition. Similarly, it is possible to construct the left adjoint to the forgetful functor from groupoids to categories, that is the free groupoid over a category.

Lauren Cote, Free groups and graphs: the Hanna Neumann theorem (pdf)

Philip Higgins, Categories and groupoids, Van Nostrand Reinhold, 1971; Reprints in Theory and Applications of Categories, No. 7 (2005) pp 1-195 (pdf available)