Frontier open-source embedding models on one OpenAI-compatible endpoint — top open models that outrank OpenAI's hosted embeddings, without the lock-in. Pay by the token, by card or USDC. Open, permanent weights — so your retrieval pipeline keeps working, no matter what.
curl https://api.justembedme.com/v1/embeddings \ -H "Authorization: Bearer $JUSTEMBEDME_KEY" \ -d '{ "model": "microsoft/harrier-oss-v1-0.6b", "input": "Embeddings that outlive their provider." }'
The best embedding models today are open source. OpenAI hasn't shipped a new one in years. Google's are excellent — but their long track record of quietly retiring products makes betting your vector index on them a risk we won't take. Open weights are the safest way to know your RAG pipeline and embeddings dataset stay robust for the long haul.
The catch is that hosting these models is expensive. So we figured: let us host it for you. The day you'd rather run your own, do it — we'll hand you the exact specs we run and what they cost us, no secrets. And if you're not there yet? No problem. Skip the monthly GPU bill.
Save yourself the cost, the ops, and the lock-in — and just embed me.
Most new RAG pipelines default to OpenAI because it's hosted and easy — and quietly give up retrieval quality and long-term control. We fix both.
We host the best open models. Start with Microsoft's Harrier 0.6B — frontier-class retrieval at just 596M parameters that beats OpenAI's hosted embeddings on the MTEB leaderboard, so you get better quality and a cheaper bill. 100+ languages, MIT-licensed. Step up to its larger 27B sibling when you need more. Open, multi-model, no lock-in.
Closed providers deprecate models and orphan your entire vector index. Ours are open-weight and permanent — the model is yours to keep, so you can run the exact same one yourself anytime and keep producing identical vectors, forever.
Discoverable and pay-per-call for autonomous agents: USDC over x402, an MCP tool, machine-readable pricing. Built for the agentic economy from day one — not just humans with dashboards.
Radical transparency: we publish the exact guide to self-host these models, and we publish what it costs us to run them. The escape hatch isn't a marketing line — it's a link.