Sounds like an on-disk database is the way to go here. Besides the advantage of persistence, you'll also probably use less memory as only the parts of the DB you use will be brought into memory. It would also mean you could share the database across multiple programs, if you were into that sort of thing. A Berkeley DB file would likely do what you need right now.

Depending on what you intend for this tool, you might want to abstract the interface to it some, so at a later point you could add more features to teh DB back-end--multiple indices, synonyms, or stuff like that.