Volcanic tremor is an important precursor to explosive eruptions and is ubiquitous across most silicic volcanic systems. Oscillations can persist for days and occur in a remarkably narrow frequency band (i.e. 0.5–7 Hz). The recently proposed magma-wagging model of Jellinek & Bercovici provides a basic explanation for the emergence and frequency evolution of tremor that is consistent with observations of many active silicic and andesitic volcanic systems. This model builds on work suggesting that the magma column rising in the volcanic conduit is surrounded by a permeable vesicular annulus of sheared bubbles. The magma-wagging model stipulates that the magma column rattles within the spring like foam of the annulus, and predicts oscillations at the range of observed tremor frequencies for a wide variety of volcanic environments. However, the viscous resistance of the magma column attenuates the oscillations and thus a forcing mechanism is required. Here we provide further development of the magma-wagging model and demonstrate that it implicitly has the requisite forcing to excite wagging behaviour. In particular, the extended model allows for gas flux through the annulus, which interacts with the wagging displacements and induces a Bernoulli effect that amplifies the oscillations. This effect leads to an instability involving growing oscillations at the lower end of the tremor frequency spectrum, and that drives the system against viscous damping of the wagging magma column. The fully non-linear model displays tremor oscillations associated with pulses in gas flux, analogous to observations of audible ‘chugging’. These oscillations also occur in clusters or envelopes that are consistent with observations of sporadic tremor envelopes. The wagging model further accurately predicts that seismic signals on opposite sides of a volcano are out of phase by approximately half a wagging or tremor period. Finally, peaks in gas flux occur at the end of the growing instability several tens of seconds after the largest tremors, which is consistent with observations of a 30- to 50-s lag between major tremor activity and maximum gas release. The extended magma-wagging model, thus, predicts tremor frequency and its evolution before and during an eruption, as well as a driving mechanism to keep the tremor excited for long periods.