how close to radioactive material to be detected I am not expert on this but we have a project to detect radioactive material, we have found an off-the-shelf devise where its sensitivity for gamma rays starts from 20 KeV. My question is if you have very small traces of a radioactive material, how close this devise should be in order to pick any gamma ray? 
 A: There are various radioactivity kinds. Take for example two kinds: alpha and gamma. While gamma are electromagnetic particles, photons, with rather big energy, it can travel a long way in open air, and even penetrate not so dense barriers like wood, glass or even concrete depending on photons energy, material structure and thickness. Of course it is absorbed in heavy metals like lead. 
On the other side alpha rays are just helium nuclei. It may be very energetic but usually they get absorbed even in air in short lengths - for example 0.5 m - in practice typical alpha get absorbed after several cm ( say 10cm) but there are exceptions. But they are very dangerous as they ionise matter with high efficiency, and they will be absorbed by the skin when you are close enough to the source. Absorption means here energy transfer, so basically skin will be just burned!
Detectors usually used one or another kind of radioemission of electrons caused by radiation absorption  in semiconductor or on various electrodes. After amplification, electric current of such electrons is measured. 
So - what is the correct answer? It depends both on energy of the radioactivity radiation and kind of the radiation, and radioactive material you are expecting to measure, detect etc. Probably you should focus on measure of gamma or beta radiation ( electrons) as they can be detected far from the source. 
A: 
My question is if you have very small traces of a radioactive material, how close this devise should be in order to pick any gamma ray?

If you don't mind waiting an unlimited amount of time to get a measurement, then the main issue to worry about is background. The count rate will fall off as $1/r^2$, so if there is no background, all you have to do is wait longer in order to get a statistically significant number of counts. But in reality there is background. If the background is, e.g., $10^9$ times stronger than your source, then you will have a very hard time subtracting out the background. You can try to shield out gamma-ray background, but there are limits to how well you can do this, often because your shielding itself will contain trace amounts of gamma-emitting isotopes.
