Pricing

Nothing important is behind a sales call.

Storage, writes and query scan are priced at what the incumbents charge. Customer-managed keys, SSO, private networking and a HIPAA BAA are included from $199 a month instead of $4,096.

Free
$0

Everything working, at prototype scale. No card.

Start free
  • 2 GB storage
  • Unlimited namespaces
  • Standard writes, 1 M queries / mo
  • Community support
Starter
$49 / mo min

Usage-based, in production, without a procurement cycle.

Start free
  • Everything in Free
  • Realtime writes (~25 ms commit)
  • Embedding + reranking stages
  • Bounded-staleness consistency
  • Email support
Most teams Scale
$199 / mo min

The security package a regulated buyer needs, at a self-serve price.

Start on Scale
  • CMEK and private networking
  • SSO / SAML and audit logs
  • HIPAA BAA, SOC 2 Type II
  • Namespace branching and pinning
  • 99.9% SLA, shared Slack channel
Enterprise
Custom

Runs inside your own GCP project. No usage premium.

Configure a deployment
  • BYOP — your project, your keys
  • VPC Service Controls, Access Transparency
  • Single-tenant cells, 99.99% SLA
  • GCP Marketplace billing
  • No 35% usage premium
Estimator

Price your real workload, not a sample.

Every line is a published list rate — ours and theirs. Change an input and watch which vendor's model breaks first.

Corpus

{{ vectorsLabel }}
million vectors = {{ storageLabel }} logical

Writes

{{ writeLabel }}
GB / month
{{ realtimeLabel }}
% realtime Realtime writes are $3/GB against $2/GB standard. No competitor offers this tier at any price.

Queries

{{ queryLabel }}
million / month
{{ scanLabel }}
MB / query

Pipeline & compliance

{{ embedLabel }}
million tokens
{{ rerankLabel }}
million candidates
Peregrine {{ tierLabel }}
{{ total }}
{{ l.name }} {{ l.detail }} {{ l.amount }}
{{ minimumLabel }} {{ total }}
Addressable estimate {{ shareUrl }}
Same workload elsewhere
{{ r.name }} {{ r.amount }}
{{ r.note }} {{ r.delta }}

Modelled from list prices published July 2026 and simplified to comparable units. Consumption-unit vendors are approximated at 2 read units per query. Your quote will differ; we will happily model it with you.

Machine-readable

Half our evaluators are agents.

A calculator behind a slider is invisible to anything that isn't a person with a mouse. So the model is published three ways: as an endpoint, as plain markdown, and as this page's own URL. Every input above has a stable name attribute and reads its value from the query string.

GET /v1/estimate

Priced by API

Unauthenticated, rate-limited, deterministic. Returns the same line items you see above plus the competitor models, so an agent can compare without scraping a chart.

/pricing.md

The rate card as text

Every unit price, tier minimum and formula in one markdown file with no JavaScript. Linked from /llms.txt so a crawler finds it before it finds this page.

?vectors=175&dims=1536

Addressable by URL

Every state of this calculator is a link. An agent constructs the query string, a human pastes it into a design doc, and both see identical numbers.

request
curl -s "https://api.peregrine.dev/v1/estimate?\
    vectors=175000000&dims=1536&\
    write_gb_month=200&realtime_share=0&\
    queries_month=50000000&avg_scan_mb=4&\
    cmek=true&compare=turbopuffer,pinecone,zilliz"
response
{
  "currency": "USD",
  "tier": "scale",
  "minimum": 199.00,
  "lines": {
    "storage": 21.50,
    "writes_standard": 400.00,
    "writes_realtime": 0.00,
    "query_scan": 0.20
  },
  "total": 421.70,
  "compare": {
    "turbopuffer": { "total": 4096.00,
      "reason": "cmek_requires_enterprise_minimum" },
    "pinecone":    { "total": 2530.00 },
    "zilliz":      { "total": 960.00 }
  },
  "as_of": "2026-07-29"
}
Rate card

Every unit, in one table.

Unit Peregrine Turbopuffer Notes
{{ row.unit }} {{ row.ours }} {{ row.theirs }} {{ row.note }}
Questions

The ones procurement always asks.

Can a bad deploy run up an unbounded bill?

No. Hard spend caps are a product feature, not a support ticket. Set a monthly ceiling per organisation or per namespace; when it is reached we throttle and alert rather than invoice.

Does this burn our committed GCP spend?

Yes. Buy through GCP Marketplace and the invoice draws down your committed spend, which usually turns a procurement cycle into a click.

What is the lock-in story?

The segment format is documented and the single-node engine is Apache-2.0. Your data is readable outside the managed service with no export step, and one-time egress of a terabyte of vectors is about $90.

Are commitments discounted?

One and three-year commitments earn CUD-style discounts on every usage line. No cliff, no overage penalty — you keep paying list above the commitment.

Which regions and clouds?

Google Cloud today, in three regions, because exploiting one cloud's fastest primitives is exactly how we beat a multi-cloud design. Cross-cloud read replicas into AWS are on the roadmap; write-path AWS cells follow demand.

What does the free tier actually cost us?

Pennies — an idle namespace on object storage is close to free, which is why it is genuinely free forever rather than a 14-day trial.

Send us your last invoice.

We will model your workload line by line and tell you honestly if the move isn't worth it.

Start free Configure a deployment