An Rspamd alternative with its own AI engine and no LLM server

Good results take a lot of configuration

Rspamd combines rules, a Bayes classifier, fuzzy hashes and an optional neural module. Each part has its own settings, and getting them to work well together on your mail takes time and Redis to store what it learns.

LLM checks are slow, costly or send mail away

Rspamd's LLM plugin is off by default. Turned on with OpenAI, it sends the text, subject, sender and links of each message it checks to OpenAI, and each request has a cost. Turned on with a local Ollama server, you run and size that server yourself, and large language models are slow on ordinary CPUs.

SpamFoo's AI engine is small and built only for reading email. It runs on the same standard CPUs as your mail server, with no GPU required, and message text isn't sent to an outside AI provider to get a verdict.

Per-user learning that starts with one click

When a user marks a message as spam or moves one out of junk, SpamFoo applies it to their mailbox right away. There is no Bayes database to train and no per-user storage to set up. Each correction also helps improve the models every SpamFoo installation uses.

Phishing protection and sorted mail

SpamFoo's Anti-Phishing protects the names your staff trust and catches domains made to look like ones you know, including letters from other alphabets. Email Classification sorts the mail people keep into Primary, Transactions, Updates and Promotions. In our research we found no documented built-in equivalent in Rspamd.

Running SpamFoo with Rspamd

Spam filters often work well together. Some people keep Rspamd running alongside SpamFoo, although most find it unnecessary once SpamFoo is filtering. SpamFoo Client runs on your mail server or on another server you control, or we can host it for you, and your mail server asks it about each message.