Low-dose-rate (LDR) brachytherapy is a standard treatment option for non-metastatic prostate cancer patients. It consists of permanent implantation of radioactive seeds in the prostate using needles. LDR seeds offer locally-confined dose deposition, sparing organs at risk (rectum, ure- thra). Selecting the implantation sites manually is a laborious task. Inverse planning algorithms aiming to achieve the optimal implantation automatically have been proposed, based on dose calculation following the dosimetry formalism of the TG43 report. TG43 over-estimates the actual deposited dose leading to sub-optimal plans. Inverse planning based on Monte Carlo Dose Kernels (MC-DK) has been proposed. Due to the computational burden of MC meth- ods, a pre-calculating step of 78h was required. Furthermore, optimization was based on indirect dose criteria. To address current planning limitations we developed ORACLE, an inverse plan- ning system using fast simulated annealing (FSA) and GPU-accelerated MC dosimetry. Using GGEMS platform, we can generate MC-DKs for a number of 424 possible implantation sites in intra-operative time (42.4s) with ≈ 2% statistic uncertainty in the prostate. The optimization is based on DVH quality criteria. Implantation configurations are evaluated iteratively by accu- mulating MC-DKs in a total dose map. Introducing a compression of the MC-DKs, allowed us to accelerate further the overall process, providing plans that satisfy the DVH criteria, given by the TG137 report, in only 15s. Comparing ORACLE-generated plans with clinical plans of 10 patients we were able to provide improved quality plans using less needles (10.5% less). A fast and efficient inverse planning system based on MC dosimetry is proposed, reducing planning dose over-estimation. Optimizing the DVH eliminates any learning-curve for the operator. In addition, delivering high quality plans with the least number of needles can reduce the induced trauma (edema), decreasing deviation between planning and post-operative dosimetry.