There are many kinds of radio receiver architectures, but a very common one is a superheterodyne, or "superhet". In a superhet, the tuned frequency can be the sum (or difference) of multiple oscillators, one of them often a VFO (an oscillator variable in frequency by changing a variable capacitor, or perhaps a PLL voltage). In a common type of single conversion superhet, the tuned frequency is the sum of an IF frequency (say 455 kHz) and the VFO frequency. So the VFO frequency (as measured by say an oscilloscope or frequency counter) is not the frequency indicated on the dial, or the carrier frequency of a tuned-to AM station.
An RTL-SDR, as typically used by many SDR applications, is actually a double or triple conversion superhet, with one VFO (actually variable) digitally synthesized in the tuner chip for heterodyning down to an IF, another digitally synthesized fixed frequency quadrature oscillator for heterodyning IF samples down to baseband IQ for transfer over a USB-port protocol, and yet another software synthesized oscillator (often variable, so yet another VFO) inside the SDR application to add or remove a tuned offset.
With SDR software that shows a spectrum and waterfall, the center of the entire available FFT waterfall is often at the hardware "VFO" frequency (which may be a single oscillator, or a sum/combination of more than one). But one can still use the SDR software demodulate a signal with a frequency above or below the center of the waterfall by IQ heterodyning with yet another offset oscillator (software VFO). The frequency indicated by the SDR software might be the arithmetic sum (or difference, depending on the signs of the frequencies) of the multiple VFOs.
The reasons for doing this are many.
One reason is so you can see an entire band, but listen to various signals within that band without shifting the entire waterfall. So one can listen to signals lower or higher in either the CW or SSB portions of a band while still looking at both across the entire waterfall, perhaps centered between them at a "VFO" frequency.
Another reason (for many RTL-SDR, mcHF/RS918, and SDRPlay devices), is that there is an IQ imbalance or offset in the initial quadrature heterodyne to IQ signals, which distorts the signal at the center of the waterfall (baseband frequency of zero), but less so for signal offset from the IQ VFO frequency. So the tuned, indicated, or "listened to" frequency is often offset by some amount (maybe a dozen kHz or so) from the VFO frequency.
In an mcHF transceiver, the offset is a configuration option, usually set at 12 kHz. So the indicated dial frequency is really 12 kHz above or below the actual VFO (a physical signal you can monitor with an oscilloscope), the quadrature signal fed to the Tayloe IQ mixer. The firmware later removes this offset by IQ software heterodyne before DSP demodulation at baseband.
In a direct sampling SDR (Elecraft K4 or Hermes Lite 2, et.al.), the first VFO is usually implemented as a digitally synthesized signal inside an FPGA, for down conversion before sample rate reduction to a sample rate more suitable for an affordable PC or embedded processor. SDR software can use this directly, or use one or more additional software oscillators for offset or multi-signal monitoring.