One signed binary. Every feature compiled in. Free to run. Install Crowkis →
← back to the Roost
guidesJune 26, 2026· 4 min read

Cache LiteLLM in your voice assistant with Crowkis

Building voice assistants on LiteLLM? Add a semantic cache so latency-sensitive, repetitive spoken queries stop costing full price.

Voice assistants built on LiteLLM share one problem: latency-sensitive, repetitive spoken queries. Each repeat is a full-price LiteLLM call for an answer you already have.

Put a semantic cache in front. Point LiteLLM's base URL at the Crowkis OpenAI-compatible gateway, or wrap the call in get-or-compute, and reworded repeats are served from cache, no rewrite of your voice assistants logic.

LiteLLM + Crowkis
base_url = "http://127.0.0.1:6380/v1"   # Crowkis gateway, semantic cache in front
In plain words: The voice assistants keep working exactly as before; the repeats just stop hitting the model.

On repetitive traffic this cuts costs up to 60-70% on repetitive workloads, and every hit carries a confidence score so reuse stays safe. Runs self-hosted with zero egress, nothing leaves your machine.