Pathology labs, medical companies and research labs count cells routinely, but counting by hand takes about a minute for an image of 50 cells, is repetitive, and varies between people. We built a deep-learning tool that returns the cell count of a microscopy image, with a box around every cell so each count can be checked. On 500 annotated brightfield images from five sessions, our YOLO26n detector, fine-tuned on our labelled cells, counts 75 held-out test images with a mean absolute error of 1.6 cells per image (R² = 0.98). The baseline, the general-purpose Cellpose-SAM model with thresholds tuned on our training images, is off by 5.6 cells (R² = 0.74). Our model is also about 63 times faster (28.6 ms vs 1.8 s per image). Applied without retraining to synthetic images, it counts small, round cells almost exactly but misses large, irregular, touching ones. Cell images vary too widely for one general-purpose model, so we expect labs to fine-tune ours on their own images.