The open-source control plane for the language your search runs on.

Your team's vocabulary drifts — features get renamed, slang piles up, acronyms collide.
Your embedding model never learned your internal names, so retrieval quietly rots.
SkeinRank detects that drift, versions your terminology, and keeps search quality under control as you scale.

Website · Quickstart · Docs · Main repo · PyPI

Retrieval quality isn't static — it decays.

A feature called checkout-v2 in January becomes payments-core in your June docs. On-call still types "the checkout thing." Three names for one reality, indexed at different times. Your embedding model never saw your internal names on the public internet, so it guesses — and as your language drifts, those guesses rot. No error, no alert, just search that's a little worse every month until someone says "the bot got dumb."

Your vocabulary is the one input to retrieval that changes constantly — and no system owns it, versions it, or tells you when it drifts.

SkeinRank is that system.

A terminology control plane that sits beside the Elasticsearch, OpenSearch, or vector DB you already run. It doesn't replace your search engine — it governs the language feeding into it.

• Drift detection — measure how far your live language has moved from the vocabulary your search relies on.
• Terminology governance — canonical terms, aliases, profiles, bindings, and immutable versioned snapshots.
• Context-aware disambiguation — pg timeout resolves differently from pg layout, by explicit rule, not a guess.
• Evidence-assisted review — validate every change against real indexed content before it ships.
• Safe rollout — blue/green alias swap, rollback, and before/after retrieval evaluation.
• Agent integration via MCP — agents can propose terminology fixes under strict RBAC, but never mutate production directly.

pip install skeinrank

import skeinrank
skeinrank.canonicalize("k8s pg timeout")   # → "kubernetes postgresql timeout"
skeinrank.canonicalize("pg layout")        # → "page layout"

Inside the monorepo:

SkeinRank is in active open-source public preview — not a hosted SaaS. Current focus: binding-aware runtime canonicalization, terminology drift detection, safe governance, Terminology-as-Code, and operator-controlled Elasticsearch/OpenSearch delivery.

Start at skeinrank.github.io · Apache-2.0 licensed.