As FvM et al have shown, the variable integrator best suits your need.
(As an aside, there are many digital synth solutions including DAC wave shaper converters (linear DAC on output of n-bit up & down counter gives triangle wave, at input freq /n. A standard up or down counter gives sawtooth with freq/n output. A since FM modulated software clock generates a sinusoidal up/down counts give a sinewave output.. random up down counts gives random signal. Any arbitrary signal can be generated using approximation at n x freq and integrate with up-down counters in h/w or s/w with DAC output DAC or analog output) You can even use logarithmic µLaw or A-Law 8 bit DACs for > 70dB dynamic range. The digital approach is useful if you need more resolution than say the PWM obvious method using double precision.) now back to the simple solution....
Theory of Operation
If you vary the speed of integration with a log pot such that its output reaches a threshold and then inverts direction of integration, you get a constant amplitude triangle wave and square wave. Sine is shaped form Triangle by log function using diode compression.
Although interesting but irrelevant, recall that a triangle and a square wave have the same Fourier magnitudes except inverted phase of harmonics. The integrator shifts the phase in the time domain by 90deg for V vs I and also shifts the polarity of harmonics in the frequency domain. from -1,1,-1,1.. to 1,-1,1,-1... ( my memory is fuzzy.. but I think that is right)
A Sine can be approximated by performing a log conversion of the peak signals of the triangle wave to shape them, as the sine and triangle have the same slope or dominant fundamental component.
For extra points you can empirically show how log of integral of square wave relates to fundamental in frequency domain.
I refer to the popular design....and leave for you to work out the values. Any 50 ohm variable gain driver can be your output. A DC level shifter is useful with a large bipolar supply.