Which application should I use?

It depends on what you have to start with. Here is a list of proposed strategies for some typical cases ....

• I have experimental phases to higher than 2.3 Å, native data to 1.8 Å or better: just run warpNtrace with default parameters ... it should just work.
• I have experimental phases to lower than 2.3 Å, native data to 1.8 Å or better: First use phase extension techniques via solvent flattening, etc. and then run warpNtrace. If the phases extend to i.e. only 3.0 Å do not extend immediately to the resolution of the native data set. Extending to 2.3 Å is enough for placement of the starting free atoms model.
• I have experimental phases for one set of data and high resolution native which is slightly non-isomorphous: A trick that often works, is to solvent flatten using the existing phases from one dataset and the amplitudes from the high resolution data set. Then go on from the solvent flattened phases ...
• I have native data to 2.0 Å and phases extended reliably to 2.0 Å but warpNtrace does not find any protein regions - but I can build a few things: Well, go ahead and build these parts. Then, before going home to sleep submit a new job using what you built as a starting model. At some point the protein part might get enough for ARP/wARP  to do the rest automatically.
• I have native data and phases to 2.5 Å or worse: Learn how to use O. After building most of the model and if data are better than 3.0 Å  at least you could try to run the molrep application which in some cases might deliver a better map. This has worked quite well in cases where i.e. the DNA part was built first, together with a few helices, and ARP/wARP delivered a better looking map.
• I have a molecular replacement model and native data better than 2.0 Å: Run the warpNtrace application.
• I have a molecular replacement model and native data worse than 2.3 Å: Run the molrep application. You may want to use side_trace to mutate easily the search model first ...
• I have a native data set to 1.1 Å resolution and a metal: Find the metal and use it as a start model for warp_solve. Then use warpNtrace.
• I have a native data set to around 1.8 Å resolution and a metal: Keep the anomalous signal separate. Most metals do have an anomalous signal even far from the edge. Use a phasing program that is designed for treating such weak signal properly (i.e. SHARP), solvent flatten and then try warpNtrace.