# Peregrine pricing

> Canonical, machine-readable rate card. Last updated 2026-07-29. All prices USD.
> Interactive version: https://peregrine.dev/pricing
> Programmatic version: `GET https://api.peregrine.dev/v1/estimate`

Peregrine is a vector and full-text search platform for search and agent retrieval,
built on object storage. Pricing is usage-based with a monthly tier minimum.

## Principle

Match the incumbents on commodity units. Do not gate security behind an enterprise
minimum. Charge a premium only where the capability has no competitive equivalent
(realtime writes, embedding, reranking).

## Unit rates

| Unit | Rate | Notes |
| --- | --- | --- |
| Storage | $0.02 / GB-month | Logical bytes stored. Near passthrough of object storage. |
| Writes — standard | $2.00 / GB | Visible in ~1 second. |
| Writes — realtime | $3.00 / GB | ~25 ms commit, synchronously durable in two zones. |
| Query scan | $1.00 / PB scanned | Bytes read to answer queries. |
| Cache pinning | $0.0008 / GB-hour | Reserved cache residency. |
| Embedding | $0.04 / 1M tokens | Optional server-side embedding on write and query. |
| Reranking | $2.00 / 1M candidates | Optional cross-encoder second stage. |
| Enterprise usage premium | none | There is no surcharge on any tier. |

## Tiers

| Tier | Monthly minimum | Included |
| --- | --- | --- |
| Free | $0 | 2 GB storage, unlimited namespaces, standard writes, 1M queries/month. |
| Starter | $49 | Realtime writes, embedding and reranking stages, bounded-staleness consistency. |
| Scale | $199 | CMEK, private networking, SSO/SAML, HIPAA BAA, SOC 2 Type II, branching, pinning, 99.9% SLA. |
| Enterprise | custom | BYOP (runs in your GCP project), VPC-SC, Access Transparency, single-tenant cells, 99.99% SLA, GCP Marketplace billing. |

The bill is `max(sum_of_usage, tier_minimum)`. Nothing is pre-purchased and no
capacity is reserved on your behalf.

## Cost formula

```
storage_gb        = vectors * dimensions * 4 / 1e9
storage_cost      = storage_gb * 0.02
write_cost        = standard_gb * 2.00 + realtime_gb * 3.00
query_scan_pb     = queries * avg_scan_bytes / 1e15
query_cost        = query_scan_pb * 1.00
embedding_cost    = embed_tokens / 1e6 * 0.04
reranking_cost    = rerank_candidates / 1e6 * 2.00

usage             = storage_cost + write_cost + query_cost
                    + embedding_cost + reranking_cost
monthly_total     = max(usage, tier_minimum)
```

## Worked example

Workload: 1 TB stored, 200 GB written per month, 50M queries per month at 4 MB
average scan, customer-managed encryption keys required.

| Line | Amount |
| --- | --- |
| Storage (1,000 GB) | $20.00 |
| Standard writes (200 GB) | $400.00 |
| Query scan (0.2 PB) | $0.20 |
| CMEK, SSO, private networking | included on Scale |
| **Total (Scale tier)** | **$420.20** |

## Competitor modelling

These figures are our model of publicly listed prices, simplified to comparable
units, as of 2026-07-29. They are not quotes and vendors change prices. Verify
before relying on them.

| Vendor | Same workload | Driver |
| --- | --- | --- |
| Peregrine (Scale) | ~$420 | CMEK included at $199 minimum. |
| Zilliz Cloud | ~$960 | Storage $0.06/GB-mo, writes $1.50/GB, ~$12 per 1M queries. |
| Pinecone | ~$2,530 | Storage $0.33/GB-mo, writes $3/GB, read units dominate at volume. |
| Turbopuffer | ~$4,096 | Unit rates match ours; CMEK requires Enterprise ($4,096 minimum plus 35% usage premium). Usage alone would be ~$567. |

The gap with Turbopuffer is not unit price — it is security gating.

## Spend control

Hard monthly spend caps per organisation and per namespace are a product feature.
On breach we throttle writes and alert; we do not continue billing. The free tier
never converts to a paid plan automatically.

## Commitments and procurement

- One and three-year commitments earn CUD-style discounts on every usage line.
  No cliff and no overage penalty — usage above the commitment bills at list.
- Available through GCP Marketplace; invoices draw down committed GCP spend.
- One-time egress of 1 TB of vectors is approximately $90. The segment format is
  documented and the single-node engine is Apache-2.0.

## Regions

`us-central1`, `europe-west4`, `asia-southeast1`. Google Cloud only today.
Cross-cloud read replicas into AWS are on the roadmap.

## Contact

pricing@peregrine.dev — send a current invoice and we will model the move line by
line, including the cases where it is not worth making.