Gamma-Ray Bursts (GRBs) were serendipitously discovered in the late 1960s by the military Vela satellites which were monitoring the Nuclear Test Ban Treaty between the US and the Soviet Union. The announcement was postponed for several years, after having ruled out a man-made origin and ascertained that they were outside the immediate solar system [1]. In a matter of a few years more than a hundred models had been proposed to explain their astrophysical origin [2], ranging from comet infalls, through stellar cataclysmic events, to events associated with supermassive black holes at the center of galaxies. The problem in making the first steps towards a theoretical understanding was that the gamma-ray instruments of the time had poor positional accuracy, transmitted to Earth only many hours after the trigger, so that only wide-field, insensitive telescopes could follow-up the bursts to look for counterparts at other wavelengths.
In the 1990s the Compton Gamma Ray Observatory (CGRO) was launched, one of whose main objectives was the detection of GRBs. The Burst and Transient Source Experiment (BATSE) onboard CGRO obtained, over a decade, the positions of ~ 3000 GRBs. This showed that they were uniformly distributed over the sky [3], indicating either an extragalactic or a `galactic-halo origin. BATSE also found that GRBs can be classified into two duration classes, short and long GRBs, with a dividing line at ~ 2 s [4].
The search for GRB counterparts at other wavelengths remained unsuccessful for almost 25 years, until in 1997 the Beppo-SAX satellite localized with greater accuracy the first long lasting X-ray afterglows [5], which in turn enabled the first optical host galaxy identification and redshift measurement [6]. The long bursts were found to be associated with galaxies where active star formation was taking place, typically at redshifts z ~ 1-2, and in some cases a supernova of type Ic was detected associated with the bursts, confirming the stellar origin of this class. The power law time decay of the light curve was also observed, in a number of cases, to exhibit a steepening after ~ 0.5-1 day, suggesting (for reasons explained below) that the emission was collimated into a jet, of typical opening half-angle ~ 5°, which eased the energy requirements. Even so, at cosmological distances this implied a total time-integrated energy output of ~ 1050 - 1051 erg. This is roughly 10-3 of a solar rest mass, emitted over tens of seconds. This is more than our Sun emits over its ten billion year lifetime, and about as much as the entire Milky Way emits over a hundred years - and that is mainly concentrated into gamma rays.
Well before the CGRO and Beppo-SAX observations, early theoretical ideas
about the origin of GRBs had converged towards an energy source provided
by the gravitational potential of a compact stellar source, the latter
being suggested by the short duration (tens of seconds) and fast
variability 
10-3 s of the
-ray
emission, using a simple causality argument R
c
t
10-100
Km. The large energies liberated in a small
volume and in a short time, as well as the observed hard spectrum
( MeV) would then
produce abundant electron-positron pairs via
photon-photon interactions, creating a hot fireball which would expand,
eventually reaching relativistic bulk velocities
[7].
Among the first stellar sources discussed which could be responsible for GRBs were binary double neutron star (DNS) mergers, or black hole-neutron star (BH-NS) mergers, whose occurrence rate as well as the expected energy liberated ~ GM2 / R appeared sufficient for powering even extragalactic GRBs [8, 9, 10, 11]. These are nowadays, the leading candidates for the short gamma-ray bursts, as shown by Swift and other observations e.g. [12]. Another candidate stellar source was the core collapse of massive stars and the accretion into the resulting black hole [13, 14]. Initially it was thought that this would result in a GRB and a failed supernova, but later observations, e.g. [15] and others, showed an unusually luminous core collapse supernova of type Ic associated with some GRBs; these supernovae have since been referred to as hypernovae. The core collapse model, referred to as a collapsar, is currently well established as the source of most long GRBs.
The predicted rate of occurrence of binary mergers and of hypernovae is sufficient to account for the number of bursts observed, even if the gamma-rays are beamed to the extent that only one event in 100-1000 is observed. (We expect less than one observable burst per million years from a typical galaxy, but the detection rate can nonetheless be of order one per day because that are so powerful that they can be detected out to the Hubble radius).