embedcache

Use case

Embedding cost control

Cut the recurring bill from re-embedding stable inputs by keeping vectors on-box in a cache that survives restarts.

The problem

A large share of embedding spend goes to re-computing vectors for inputs that never changed — the same FAQ, the same docs, the same product descriptions, over and over.

How embedcache helps

  1. 1

    Route embedding requests through embedcache so identical inputs resolve from the cache.

  2. 2

    Because the cache is content-addressed, deduplication is automatic — no manual keys to manage.

  3. 3

    Move from a per-token operating expense to a one-time local compute cost plus a small SQLite file.

  4. 4

    Measure the effect with the cache indicator returned per input on /v1/embed.

Features involved