Levenshtein distance is the number of operations/edits you need to do, to change one word into the other. The regularly used weighting is:
1 for substitution (example: 1 wrong letter typed)
1 for deletion (example: 1 letter too much)
1 for addition (example: 1 letter missing)
So you could say you allow 1 typo in the word, then for red stuff like this would be allowed:
reed, ret, rad, rwd, rd, ...
Of course you could make it reliant on word lenght, so for example forbid typos in short words, but accept one or two in longer words.
Word suggestion is something different and would use a dictionary, preferably one which knows the frequency of words in common usage, to suggest the most frequently used word(s) that begin with the typed letters.
Would it not be a problem in your game, if autocompletion is too strong? (it could get weird for the players if all they do is write out words half-way and then the next task already comes up)