We are in need of a web proxy with a few custom features.

The proxy should be able to do http 1.0 and 1.1 requests over http and https.

It should transform all text/html and xml and plain/text content to uf8.
It should parse and rebuild all html/sgml/svg/$ml to clean valid xml (while keeping the doctype for vml/svg etc)

It should do multiple concurrent connections.
It should be able to honor keepalive by pooling requests to the same server.

For any requests made it can check with a separate webservice if the request should be halted, diverted or passed trough without parsing. (parsing would involve buffering, which we don't want on large non/$ml files.)

We think Libevent + Beatifull soup / hubhub or some other clean fast html parser should to the trick.

Detailed specs for the filtering webservice are available.