Solar activity in the past millennia can only be reconstructed from cosmogenic radionuclide records in terrestrial archives. However, because of the diversity of the proxy archives, it is difficult to build a homogeneous reconstruction. Here we provide a new consistent multiproxy reconstruction of the solar activity over the last 9000 years, using available long-span datasets of 10Be and 14C in terrestrial archives. A new method, based on a Bayesian approach, was applied for the first time to solar activity reconstruction. A Monte Carlo search for the most probable value of the modulation potential was performed to match data from different datasets for a given time. We used six 10Be series from Greenland and Antarctica, and the global 14C production series. The 10Be series were resampled to match wiggles related to the grand minima in the 14C reference dataset. The GRIP and the EDML 10Be series diverge from each other during the second half of the Holocene, while the 14C series lies between them. A likely reason for this is the insufficiently precise beryllium transport and deposition model for Greenland. A slow 6-millennia variability with lows at ca. 5500 BC and 1500 AD of solar activity is found. Two components of solar activity can be statistically distinguished: the main 'normal' component and a component corresponding to grand minima. A possible existence of a component representing grand maxima is indicated, but it cannot be separated from the main component in a statistically significant manner. A new consistent reconstruction of solar activity over the last nine millennia is presented with the most probable values of decadal sunspot numbers and their realistic uncertainties. Independent components of solar activity corresponding to the main moderate activity and the grand-minimum state are identified; they may be related to different operation modes of the dynamo.