Hi Marc, I didn't run any real performance test with it myself yet, but what I'm understood from other Lucene users is that with the finite state machine approach for auto-suggestion, the "index" (more appropriately the state machine) isn't suited for real-time rebuilding. We'll have to create a method which can rebuild it on demand - similarly to the MassIndexer. I expect it to take much less time as it wouldn't need to literally reload all of your database, but still it might take some minutes so not something that we should do on a per-commit operation. Maybe we could do it in background, on a fixed periodicity. I'd prefer to delegate such cron jobs to user's coding responsbility (which could be our first implementation step), but there might be some additional optimisations which we could do when controlling the process. I'm thinking of: - disable it when running the MassIndexer - automatically run it at end of MassIndexer - disable a timed event if there have been no changes during the period Would you see a use for it be "guaranteed" to be always in synch, synchronously, even if it took you - say - 2 minutes of intense CPU and IO operations to build? Sanne On 23 February 2015 at 21:28, Marc Schipperheyn <m.schipperheyn(a)gmail.com&gt; wrote: