“Local” is a word a lot of tools use loosely. Here is what it means for Axon, concretely.
The documents stay where they are
You point Axon at folders that already exist on your disk. It indexes them in place. Nothing is copied into an app container, a cloud bucket, or a “workspace” — which also means there is never a second copy quietly drifting out of step with the one you actually edit. Per-project ignore rules decide what gets indexed at all.
The index is local too
Parsing, chunking, embeddings and the vector store are all computed and kept on your machine. Answering a question is a sequence of reads against files on your own disk and a model running on your own processor. There is no API key to configure and no outbound request to make, so Axon works on a plane, in a client's facility that blocks outside tools, and behind a firewall that allowlists nothing.
The model is local too
Axon answers with the model it ships with, or with your own Ollama server if you run one. It does not send your question or the retrieved passages to a cloud model — there is no setting that does.
Sharing, when you turn it on
Sharing a project (on Pro) gives a colleague's machine on your local network a copy — read-only unless you grant write access — and keeps it in sync, directly between the two machines and without going through anyone's servers. When both run the same models the search index comes across too, so nothing is re-indexed. Nothing is shared until you start it.