tool · pip install awembed

awembed

Train an embedding model that knows your corpus, and prove it beats the big one.

The Aither World is an operating system for agents — a Linux you can hand to one, the runtimes it works in, and the tools it works with. awembed is one of its 55 bricks — each installs on its own, runs offline, and needs no account.

The problem

Every agent stack searches your code with an embedding model trained on someone else's. It is right about two thirds of the time on a corpus it never saw, and nothing in the stack measures that. Distilling a small student on your own corpus -- the big model's margins plus your labels -- beats the big model, and the eval that proves it is the part people skip.

instead of trusting

a general-purpose embedder that has never seen your code

you check

a held-out split of whole directories, scored teacher vs student vs int8

Adopt it

pip install awembed

Point it at one repo and get a 0.6B embedder that ranks your directories better than the 7B one it learned from, with the eval that proves it.

Pairs with

awdk awgraph awm awfind awrecurse awrepl awsh awskills

Pairing is composition, never dependency — awembed installs and runs on its own.

The rest of the Aither World

Every brick below is live, drawn from that repository's own published manifest.