Efficient computation of the angular bispectrum is an essential part of modelling large-scale structure observations, but it still remains an extremely challenging task. In this work, we compute the tree-level, unequal-time angular bispectrum in both real and redshift space. By deriving full-sky results, we show that the bispectrum can be expressed as a sum of products of two angular power spectra, enabling the use of our recently developed flat-sky approximation to enhance computational efficiency significantly. This flat-sky formalism preserves key line-of-sight mode information while discarding extraneous full-sky contributions. We validate our approach by comparing it with direct full-sky integration, finding excellent agreement across a wide range of scales and redshifts for all bispectrum configurations. At redshift $z = 1$, we achieve sub-percent agreement (for multipoles $\ell \gtrsim 5$) between full-sky and flat-sky results for equilateral, squeezed, and folded configurations, using narrow Gaussian radial window functions ($\sigma_z = 0.01$) in both equal-time and unequal-time scenarios. On small scales, where direct full-sky integration becomes computationally prohibitive, our results align with the Limber approximation (where applicable), confirming the robustness and accuracy of our implementation. To facilitate future studies, we provide a \texttt{Python} implementation of our results, which is publicly available on \texttt{GitHub}.