Postgres for AI workloads.
Without the platform tax.
Dedicated PostgreSQL on NVMe with tuned pgvector. Fixed monthly pricing from $15/month. Migration help from Supabase, Neon, and Pinecone — included.
Free tier · No credit card · pgvector ready in 60 seconds
Your pgvector setup is probably costing you
Vector workloads don't fit neatly into general-purpose Postgres pricing. Most teams we talk to are losing money, losing sleep, or both.
Your Supabase bill is climbing
Compute add-ons, storage, egress, and vector index memory keep adding up. The pgvector workload you run for one feature is now your biggest Supabase line item.
Neon cold starts are killing your RAG latency
Serverless Postgres is elegant until the first query of the day takes 2 seconds. Usage-based pricing feels cheap until it isn't.
Pinecone is expensive and you want SQL back
You're paying for a dedicated vector database while your app data lives in Postgres. One more service, one more bill, one more copy of the data to keep in sync.
Self-hosting pgvector became a second job
You chose a VPS to save money. Now you maintain Patroni, backups, PITR, upgrades, and HNSW tuning at 2am instead of shipping.
Free tier, no credit card. Or get a workload review →
Migrating from Supabase, Neon, or Pinecone?
We'll plan the move. Free.
Already running pgvector somewhere and want a sanity check before switching? Send us your row counts, dimensions, QPS targets, and what's hurting today. You'll get back a plan recommendation, expected performance, side-by-side cost vs. your current provider, and a migration path — usually within 48 hours. If Rivestack isn't a better fit, we'll say so in the same reply.
Rivestack vs. what you have today
Same dedicated compute. Fraction of the cost. None of the platform bloat.
| Rivestack | Supabase | Neon | |
|---|---|---|---|
| Monthly cost | $15 | ~$105 | ~$69+ |
| Compute | 2 vCPU · 4 GB | Shared · 1 GB* | Serverless |
| Storage | NVMe | gp3 SSD | Cloud SSD |
| pgvector | Tuned (HNSW) | Extension only | Extension only |
| Backups | Daily + 14d PITR | Daily | Yes |
| Vector perf | ~1,000 QPS · 3.7ms p50 | Shared SSD | Serverless |
| Terraform | ✓ | ✓ | ✗ |
Supabase and Neon are excellent platforms with broader feature sets — auth, storage, edge functions, realtime. If you need those, use them. Rivestack is the focused choice for teams who want managed Postgres + pgvector without the platform tax.
Free tier, no credit card. Already on Supabase or Neon? Get a workload review →
Try it yourself
ask.rivestack.io runs on a real Rivestack cluster. Every query hits pgvector on NVMe.
Semantic search, built into your database
pgvector + HNSW on your own dedicated NVMe Postgres. Embeddings are generated for you and auto-embed on insert — load the sample dataset and search by meaning in seconds. No second database to run or sync.
- One-click sample dataset. Load it and run a real semantic query in seconds, before you write a line of code.
- Embeddings generated for you. Text auto-embeds into your vector column on insert via OpenAI text-embedding-3-small (1536-d) — you never hand-build a vector.
- Numbers measured by Postgres. Query time comes from EXPLAIN ANALYZE, with the HNSW index hit and warm p50/p95 shown — what you measure is what you get.
Benchmarked with pgvector-bench (clustered 1536-d data, HNSW m=16 / ef_construction=64, recall@10 0.93): ~1,000 QPS at a 3.7 ms p50 on a $15 node (4 clients), scaling to ~1,570 QPS at ~10 ms p50 (16 clients); ~2,500 QPS on a Scale node. Throughput and latency rise together — you don't get peak QPS at the lowest p50. In-DB query time is Postgres execution time, not end-to-end latency. Need a dedicated vector DB at your scale? We'll tell you.
Works with your stack
It's just PostgreSQL. Use your existing driver.
import psycopg2
conn = psycopg2.connect("postgresql://appuser@db-7q9m2p.eu.rivestack.io/primary")
cur = conn.cursor()
# pgvector is already enabled — just create your table
cur.execute("""CREATE TABLE docs (
id BIGSERIAL PRIMARY KEY,
text TEXT NOT NULL,
emb vector(1536)
)""")
# Insert a document with its embedding
cur.execute(
"INSERT INTO docs (text, emb) VALUES (%s, %s::vector)",
("deploy LLMs in production", embedding)
)
# Nearest-neighbor search in < 4ms
cur.execute(
"SELECT text, emb <=> %s::vector AS dist FROM docs ORDER BY dist LIMIT 5",
(query_embedding,)
)
for text, dist in cur.fetchall():
print(f"{dist:.4f} {text}")Go, Node.js, Java, Rust, Ruby, .NET — full snippets in our docs.
Edit your data like a spreadsheet
Browse, filter, and edit rows straight from the dashboard — no SQL required. Inline filters, keyset pagination that stays fast on big tables, and a side-panel editor for whole rows. Works on shared, Solo, and HA databases.
- Browse rows in a fast, sortable grid
- Filter by any column — =, contains, ranges, is empty
- Edit a whole row in a side panel, or insert and delete rows
- No SQL needed — but the SQL Editor is one click away
Provision with Terraform
Manage Rivestack like the rest of your infra. Create, scale, and destroy clusters with one file and one command — no clicking through dashboards.
Built in France · Infrastructure in EU, US-East and Singapore
What early users say
“RiveStack does care about you, no matter if you are on a paid or free plan…the guys do their best to make you happy, and your project succeed. Their support is as lightning fast as their database.”
“RiveStack is lightning fast, Postgres is online in seconds with pgvector ready for AI. The SQL tooling is solid and the AI query helper nails prompts…truly production-ready”
“delivers impressively fast provisioning, databases are prd/dev ready to connect, saving me setup time. SQL tooling is solid, and the AI-assisted query generation responded accurately to my prompts.”
Fixed monthly pricing
One price per node. No per-query billing, no egress surcharges, no vector-count overage. The bill on day 30 is the same as the quote on day 1.
Prices shown are EU Central, our lowest region. US-East and Singapore are priced higher — see full pricing.
Migrating from another provider? Get a free workload review →
Production-ready dedicated PostgreSQL with NVMe storage, automated backups, and monitoring. Add nodes for HA.
- 2 vCPU · 4 GB RAM per node
- 55 GB NVMe storage
- ~300K vectors (1536d)
- pgvector tuned (HNSW)
- HA ready (add nodes)
- Daily backups + 14d PITR
- Monitoring dashboard
- SSL encrypted
- Terraform provider
High-performance dedicated PostgreSQL for demanding workloads. Add nodes for automatic failover.
- 8 vCPU · 16 GB RAM per node
- 295 GB NVMe storage
- ~1M vectors (1536d)
- pgvector tuned (HNSW)
- HA ready (add nodes)
- Daily backups + 14d PITR
- Priority support
- Terraform provider
- Custom PostgreSQL config
The real questions
What buyers actually ask before moving a pgvector workload.
For pgvector-heavy workloads on dedicated compute, yes — almost always. A 2 vCPU / 4 GB Supabase compute add-on runs ~$105/month. The equivalent dedicated VM on Rivestack is $15/month (Solo), on NVMe instead of gp3. Send us your current plan and row counts and we'll give you a realistic number, not a marketing one. If Rivestack is not cheaper for your workload, we'll say so.
Rivestack runs dedicated Postgres on always-on NVMe nodes. No scale-to-zero, no cold starts, no cache warm-up before your first query of the day. If p95 latency and cost unpredictability are what's hurting you on Neon, that's exactly the problem we solve.
It's a real migration. pgvector with HNSW, tuned correctly on NVMe, handles the vast majority of Pinecone workloads at a fraction of the cost — and you get SQL joins, filters, and transactions in the same database as your app data. Submit your workload shape (rows, dimensions, QPS, filters) and we'll tell you whether it fits. If you actually need a specialized vector DB at your scale, we'll tell you that too.
It's standard PostgreSQL, so yes. For most databases, a pg_dump / pg_restore takes minutes. For larger or always-on workloads, we help with logical replication cutover. Migration help is included in the workload review — we don't charge extra for it.
Rivestack is a French company and our EU region runs entirely within the European Union. Your data never leaves EU territory unless you explicitly choose our US-East region. DPAs are available on request. If Europe-friendly hosting is a hard requirement for your team, this is handled by default.
Fixed monthly per node. No per-query billing, no egress surcharges, no vector-count overage. You pick the plan, you know the bill. If you need more capacity, you resize the node — you don't open an invoice and find a 4× spike.
