Splat-SLAM is a novel RGB-only SLAM system that utilizes a dense 3D Gaussian map representation and globally optimized tracking to achieve high-quality scene reconstruction and rendering. The system combines frame-to-frame tracking using recurrent dense optical flow with a deformable 3D Gaussian map, which adapts to keyframe poses and depth updates through online loop closure and global bundle adjustment (BA). To improve surface reconstruction accuracy, the system leverages a proxy depth map that combines multi-view depth estimation and monocular depth. The proxy depth map enhances rendering and reconstruction quality by refining the depth updates in inaccurate areas. Experiments on the Replica, TUM-RGBD, and ScanNet datasets demonstrate that Splat-SLAM achieves superior or comparable performance to existing RGB-only SLAM methods in terms of tracking accuracy, mapping quality, and rendering fidelity, while maintaining small map sizes and fast runtimes. The source code for Splat-SLAM is available at <https://github.com/eriksandstroem/Splat-SLAM>.Splat-SLAM is a novel RGB-only SLAM system that utilizes a dense 3D Gaussian map representation and globally optimized tracking to achieve high-quality scene reconstruction and rendering. The system combines frame-to-frame tracking using recurrent dense optical flow with a deformable 3D Gaussian map, which adapts to keyframe poses and depth updates through online loop closure and global bundle adjustment (BA). To improve surface reconstruction accuracy, the system leverages a proxy depth map that combines multi-view depth estimation and monocular depth. The proxy depth map enhances rendering and reconstruction quality by refining the depth updates in inaccurate areas. Experiments on the Replica, TUM-RGBD, and ScanNet datasets demonstrate that Splat-SLAM achieves superior or comparable performance to existing RGB-only SLAM methods in terms of tracking accuracy, mapping quality, and rendering fidelity, while maintaining small map sizes and fast runtimes. The source code for Splat-SLAM is available at <https://github.com/eriksandstroem/Splat-SLAM>.