Embeddings that keep intent apart from everything else.

WHY RESONANT

Embeddings were never trained to preserve intent.

Traditional embedding models optimize for semantic similarity. Complaints, questions, and praise about the same topic often end up in nearly identical neighborhoods.

Intent, urgency, sentiment, and sarcasm get averaged away long before retrieval ever sees them.

Resonant preserves those signals as independently recoverable information inside a single embedding, so downstream systems can retrieve, filter, and reason over both meaning and intent without a separate model.

WHEN INTENT CHANGES THE OUTCOME

When intent changes the outcome.

Semantic similarity is enough until systems need to make decisions. Routing, recommendations, memory, and agent workflows all depend on understanding what users are trying to accomplish.

BENCHMARKS

Competitive on semantics. Unmatched on intent.

Resonant matches existing models on semantic similarity while preserving intent as an independently recoverable signal. These are measured results from our evaluation suite.

Formal and casual versions of identical content are recognized as having different intent. Same intent expressed differently is recognized as equivalent.

IntentBench

Existing leaderboards answer one question: Did retrieval find the right topic?

IntentBench answers a different one: Did retrieval preserve what the user wanted?

Open benchmark. Details at general availability.

MIGRATION

Replace your embedding endpoint, not your stack.

Resonant uses the same API shape as existing embedding providers.

QUICKSTART

One API call. One embedding. Two independent signals.

Compatible with the OpenAI embeddings API. Set return_intent=True to receive separable intent axes alongside every embedding.

from resonant import Resonant

client = Resonant(api_key="rsn-...")

r = client.embed(
    input="Cancel my plan. Now.",
    model="resonant-embed-01",
    return_intent=True,
)

r.embedding   # 1536-d semantic + intent vector
r.intent      # {"urgency":0.94,"sentiment":-0.71}

WHY NOW

LLMs made intent more valuable. Embeddings never caught up.

AI systems no longer stop at retrieval. They route requests, trigger workflows, decide when to ask follow-up questions, and maintain long-term memory.

Those decisions depend on intent, not just semantic similarity.

Embeddings became the foundation of AI infrastructure. They were never optimized for the signal that now matters most.

PRICING

Drop-in pricing for intent-aware embeddings.