# Vector memory in Qdrant

> HAPP stores conversation embeddings in your Qdrant cluster, so every assistant remembers context across sessions and queries your knowledge base in real time.

- **Integration:** Qdrant (Developers & data)
- **Page:** https://happ.tools/integrations/qdrant
- **Setup guide:** https://docs.happ.tools/ — step-by-step connection instructions, error reference and connection management live in the docs, not on this page.
- **Action:** [Connect Qdrant](https://my.happ.tools/)

## What Qdrant gives your assistant

*Features*

Long-term memory, RAG retrieval, and millisecond vector search — your data lives in your Qdrant, HAPP just reads and writes.

- **Assistant remembers across sessions** — Every chat, call, and form submission is embedded and stored — the assistant recalls relevant past interactions when a customer returns.
- **Ground answers in your knowledge base** — Upload product docs, FAQs, policies — the assistant retrieves the right passage before answering, citing the source.
- **Vector queries under 50 ms** — Qdrant's Rust core keeps similarity search sub-50ms even at millions of vectors — no perceptible pause in live conversations.
- **Your data, your infrastructure** — Qdrant Cloud, Docker, Kubernetes — HAPP works the same way. Sensitive data never leaves your perimeter if you don't want it to.
- **OpenAI, Cohere, custom — your choice** — Configure which embedding model HAPP uses. Switch later without touching code — your collection migrates automatically.
- **Millions of vectors, isolated per tenant** — Sharding, replication, and HNSW indexing handle production workloads — from 10K conversations to billions of points. One cluster carries many isolated collections, so dev/prod or client-A/client-B never share an index.

## What you need to connect

*Before you start*

A short checklist — a free-tier cluster is enough to start.

1. **Qdrant Cloud account** — Sign up at cloud.qdrant.io — the free tier is enough to start. Already run Qdrant elsewhere? Self-hosted works too.
2. **A cluster created** — In Qdrant Cloud click Create Cluster — pick a region and the free plan. It takes about 30 seconds to provision.
3. **Endpoint and API key** — Open your cluster — the Endpoint field at the top is the URL you paste into HAPP. Create the API key in the API Keys tab.
4. **HAPP account with assistant memory** — Active HAPP workspace at my.happ.tools with at least one assistant. Vector memory lets it recall context across conversations.

## Frequently asked questions

*FAQ*

Answers to the most common questions about the Qdrant integration with HAPP.

### Is my data private — does it leave my Qdrant cluster?

No. Vectors live in your cluster — HAPP only reads and writes through your API key. Self-hosted Qdrant: nothing leaves your perimeter. Qdrant Cloud: data sits in your tenant under your terms with Qdrant Solutions.

### Which embedding model does HAPP use?

OpenAI text-embedding-3-small by default (1536 dimensions). You can switch to text-embedding-3-large, Cohere, or a self-hosted sentence-transformer in HAPP → Integrations → Qdrant → Settings.

### Can I switch embedding models later?

Yes. HAPP re-embeds historical conversations into a new collection in the background — no downtime, no manual migration. The old collection is kept for 30 days before deletion.

### Self-hosted vs Qdrant Cloud — which should I pick?

Cloud is the fast path: managed cluster, HTTPS endpoint, about 5 minutes to running. Self-hosted (Docker / K8s) is for compliance reasons (data residency, air-gapped) or when you already run vector workloads.

### How many vectors before performance degrades?

Qdrant comfortably handles 10M+ vectors per collection on a single node, holding sub-50ms similarity search across that range. Beyond it, sharding kicks in automatically on Cloud and manually on self-hosted.

### Can I connect multiple Qdrant clusters to HAPP?

Yes. Each cluster is a separate integration with its own URL and API key. Useful for splitting dev/prod or isolating clients. Plans differ by the number of connected clusters — see HAPP → Billing.

### What happens to vectors if I disconnect the integration?

All vectors stay in your Qdrant cluster untouched. The assistant simply stops writing new ones and stops querying for context until you reconnect — no data is deleted on either side.

## Ready to give your assistant a memory?

Point HAPP at a Qdrant cluster you control — every conversation becomes recallable context.

[Connect Qdrant](https://my.happ.tools/)