Skill search for AI agents.

Agents describe what they need in plain language. Runics finds the right skill — trust-scored, version-ranked — in under 50ms.

findSkill(

Press Enter to search the live registry

Searching

No match — Runics would hand off to Forge to generate a skill.

How Runics Works

The core problem is a vocabulary gap. An AI agent searching for "make sure we're not shipping GPL code" needs to find cargo-deny, whose description says "check Rust crate licenses." A single embedding yields ~0.58 similarity. Not good enough.

Runics solves this in two layers. At index time, each skill is enriched with an agent-optimized summary and alternate query phrasings — the vocabulary work is done once. At query time, a confidence gate routes results through three tiers:

Tier	When	Latency	LLM Cost
1 — High	Strong match, clear gap	~50ms	$0
2 — Medium	Moderate confidence	~50ms + async	Minimal
3 — Low	Weak match	500–1000ms	Full LLM call

~70% of queries resolve at Tier 1 — vector search only, zero LLM cost. Every result carries a trust score from Cognium (our verification engine) and is filtered by status: revoked and degraded skills never surface.

Get Started

npm install @runics/client

import { RunicsClient } from '@runics/client'

const runics = new RunicsClient({ apiKey: process.env.RUNICS_KEY })
const { results } = await runics.findSkill("check license compliance")