Plans & pricing

Pricing you can actually predict

Vectros is credit-based, pay-as-you-go: one credit is a penny ($0.01), and you spend credits on the work you do — with a generous free read allowance on every plan, so reading back what you put in stays effectively free at the volumes most apps run.

Inference is billed against a separate, prepaid balance — a dollar wallet you top up — so model spend stays visibly distinct from platform usage.

The credit model

What makes the bill predictable.

The thing that makes pricing predictable is what doesn’t cost anything and what scales with your data, not the platform’s.

A generous free read allowance

Every plan includes a large monthly allowance of reads — querying, fetching, and listing your own records and documents — that grows with your tier. Read-heavy AI and RAG apps stay effectively free within it; past it, a small pay-as-you-go rate, never a meter ticking from the first read.

Pay for what you turn on

Turn off embedding on a record type and there’s no vector cost on it. Turn audit history off for a high-volume, low-value type and you don’t pay to retain its history. Nothing compliance-shaped is billed unless you switch it on.

Cost tracks your data

Each customer gets an isolated index, so you never pay for other tenants’ scale. Your bill follows your own writes, storage, and searches — not the system’s size.

No surprise meter

Usage is visible in the developer portal as you go — credits drawn, by operation — so the invoice is never a mystery.

The credit schedule

What each operation costs.

This is the metering model, not a tier price sheet. Because one credit is a penny, this schedule is all you need to estimate your own platform cost — multiply credits by $0.01. Reads draw a generous per-plan allowance, then a small pay-as-you-go rate.

Records & identity

OperationRateNotes
Write — base5 credits / 1,000 writesRecords and identity (users / orgs / clients). The same base applies to document updates.
+ Equality index+1 credit / 1,000 writesPer index you can look the item up by — each ownership link, external ID, and equality lookup field set on the write.
+ Range index+3 credits / 1,000 writesPer sortable / range-queryable lookup field you enable — the added cost of range and prefix queries over an equality lookup.

Reads & data-out

OperationRateNotes
Read — get, list, queryFree allowance, then 1 credit / 500 readsA large monthly read allowance is included on every plan (it grows by tier); beyond it, this small pay-as-you-go rate. Read-heavy workloads stay effectively free at the volumes they target.
Data-out (egress)Free allowance, then 20 credits / GBAll bytes leaving the platform — every read response and document download alike. A per-plan GB allowance is included; beyond it, $0.20/GB.

Documents

OperationRateNotes
Ingest (create)1 credit / 50 documentsThe write base is folded in; the per-index lookup charges (above) still apply to each document.

Search

OperationRateNotes
Indexing (at ingest)1 / 2 / 3 credits per MBText / semantic / hybrid — charged once when content is indexed; semantic and hybrid include running the embedding model on your content.
Text query1 credit / 200 queriesBM25 keyword search — a flat rate that doesn’t grow with your index size.
Vector & hybrid queryFrom 1 credit / 175 queriesVector (semantic) similarity and combined text + vector search (the recommended default), priced by index size — a smaller index costs less to search. Up to 100K vectors: 1 credit / 175 queries; 100K–1M: 1 credit / 100. Larger indexes scale with their cost: 1M–2.5M: 1 / 75; 2.5M–5M: 1 / 50; above 5M: custom.

Per LLM call

OperationRateNotes
Inference (platform fee)1 credit / 5 callsA small metering fee per LLM call. Token cost is separate — see the pass-through table below.

Storage (monthly)

What’s storedRate
Records (structured + indexes + lookup rows)5 credits / 100 MB·mo
Documents & identity (structured, in DynamoDB)5 credits / 100 MB·mo
Large payloads & document bodies (in S3)1 credit / 100 MB·mo
Search index — text / semantic1 credit / 100 MB·mo
Search index — hybrid2 credits / 100 MB·mo
Audit / version history1 credit / 100 MB·mo

The write base covers up to 2 KB of inline data per write; larger inline payloads add a small per-KB size surcharge, but most large payloads are stored in S3 and bill on the cheaper storage line instead. Storage is metered on what you actually hold and billed at the end of each cycle. Per-tier monthly credit allowances and subscription prices are provided when we provision your plan.

Worked examples

What real workloads cost.

Four illustrative workloads, costed straight from the schedule above. Platform credits are the metered work × $0.01; inference is billed separately against a prepaid balance at the pass-through rate below.

Agentic memory

A durable memory layer for an AI assistant. Your own model runs inference — Vectros stores and recalls the memory.

5,000 memory writes · 2 indexes35 cr
~10 MB / ~5K vectors indexed (hybrid)30 cr
5,000 hybrid recalls (≤100K-vector index)29 cr
~20 MB stored1 cr
Platform credits~95~$0.95 / mo
Inference$0 — your own model

Indie RAG SaaS

A live RAG product — user uploads, hybrid search, AI answers (Sonnet).

25,000 record writes · 2 indexes175 cr
2,000 document ingests40 cr
~40 MB / ~20K vectors indexed (hybrid)120 cr
30,000 hybrid searches (≤100K-vector index)171 cr
2,000 RAG calls (platform fee)400 cr
~350 MB stored~11 cr
Platform credits~917~$9.17 / mo
Inference~$24 / mo (Sonnet, pass-through)

Clinical knowledge base

PHI-handling clinical search with AI summaries, on the BAA-eligible substrate — with a range index for date-bounded queries.

80,000 record writes · 2 indexes + 1 range800 cr
10,000 document ingests200 cr
~200 MB / ~100K vectors indexed (hybrid)600 cr
60,000 hybrid searches (100K–1M-vector index)600 cr
8,000 RAG calls (platform fee)1,600 cr
~1.3 GB stored (incl. audit history)~73 cr
Platform credits~3,873~$38.73 / mo
Inference~$45 / mo (pass-through, model-dependent)

Compliance document pipeline

High-volume ingest, search, and records for compliance — no LLM in the loop.

200,000 record writes · 2 indexes1,400 cr
60,000 document ingests1,200 cr
~1 GB / ~500K vectors indexed (hybrid)3,000 cr
200,000 hybrid searches (100K–1M-vector index)2,000 cr
~9 GB stored~270 cr
Platform credits~7,870~$78.70 / mo
InferenceNone

Illustrative, synthetic workloads — your usage will differ. Platform-credit line items sum to the total at $0.01 per credit; “indexes” is the number of lookup paths maintained per write, and examples assume payloads stay within the 2 KB inline base (no size surcharge), with large document bodies stored in S3. Storage lines blend the per-surface storage rates above (DynamoDB, S3, and audit history). Inference figures are examples for the stated model and typical token sizes.

Why it’s cheaper

The cost advantage is architecture.

The bill is lower for structural reasons, not a discount. Each of these is a property of how the platform is built — and several of them are why this costs less than running a dedicated vector database.

Reads run on a generous allowance

A large included read allowance on every plan keeps read-heavy apps (the typical AI and RAG shape) effectively free — then transparent pay-as-you-go past it, instead of a meter on every read from the first one. Your cost tracks the work you do, not reading back what you already put in.

No idle-index tax

Search is S3-backed and stateless, not an always-on in-memory vector index. You pay to index once at ingest (which includes running the embedding model on your content), then to query — never a standing fee to keep idle, rarely-queried data ready, the way a RAM-resident vector database bills its index around the clock whether you touch it or not.

Embedding is opt-in, per type

Turn embedding off on a record type and there’s no vector cost on it. You pay for the capabilities you actually switch on, per type.

You don’t pay for others’ scale

Each customer gets an isolated index. Your bill follows your own writes, storage, and searches — not the size of the platform or other tenants.

Pass-through inference

Token cost passes through at AWS Bedrock’s rate with zero platform markup. The only platform charge on an LLM call is a small per-call metering fee (1 credit per 5 calls).

For the full trade-off picture against building it yourself, see how Vectros compares.

Inference pass-through

Model tokens at cost, zero markup.

Token cost passes straight through at the underlying per-million-token rate. Vectros makes no margin on tokens — the only platform charge on an LLM call is the small per-call metering fee in the schedule above.

ModelAvailable onInput / M tokensOutput / M tokens
Claude Haiku 4.5All plans$1.00$5.00
Claude Sonnet 4.6Starter and up$3.00$15.00
Claude Opus 4.8Pro and up$5.00$25.00
Rates are pass-through at the underlying model cost — Vectros adds no markup on tokens. Inference is served from US-region infrastructure by default, which adds roughly a 10% regional cost; that premium is also passed through at no margin, never a Vectros markup, and the default can be adjusted via account management. The deployed GET /v1/models endpoint is always the live source of truth for the rates your plan can call.

The plans

Five tiers on one identical architecture.

The security substrate — per-customer isolation, encryption, scope enforcement, edge protection — is the same on every plan, including Free. Tiers differ in how much you can do, which models you can reach, and which governance controls you can turn on — not in how isolated or protected your data is.

PlanWho it’s forModel accessComplianceInference billing
FreeKick the tires; side projects; a first prototypeHaikuFull isolation + the always-on substratePrepaid balance
StarterA shipping side project or early productHaiku · Sonnet+ the substratePrepaid balance
ProMost popularA real product with customersHaiku · Sonnet · OpusBAA availablePrepaid balance
ScaleHigher volume, more headroomHaiku · Sonnet · OpusBAA availablePrepaid balance
EnterpriseContracted, regulated, or high-volumeNegotiated model accessBAA availableUsage-based billing
  • Higher tiers raise your monthly credit allowance and your throughput — the headroom grows with the plan. Exact allowances and prices are provided when we provision your plan.
  • A Business Associate Agreement is available from Pro up. The architecture that makes a BAA possible — isolation, audit, redact-at-write, in-perimeter inference on the partner data plane — is identical on every tier; the BAA itself is a Pro-and-above contractual step. Specifics are available under NDA.
  • Model access steps up by tier: Haiku on every plan, Sonnet from Starter, Opus from Pro. The deployed GET /v1/models is always the source of truth for what your plan can call.
  • Enterprise is Custom-priced and adds usage-based inference billing and negotiated model access for contracted workloads.

During the invite-only preview

These are the plans you’ll choose from — but Vectros is in an invite-only 0.x preview today, so there’s no self-serve signup yet. Request early access, tell us what you’re building, and we’ll get you in and provision a plan that fits. Self-serve is on the way as we open access — the pay-for-what-you-turn-on model you see here is the one you’ll keep.

Evaluating for a regulated workload? BAA coverage, attestation status, and the exact scope of the in-perimeter inference path are available under NDA.

Exact per-tier prices, monthly credit allowances, and rate limits are provided in the developer portal when we provision your plan — never duplicated here, so they can’t drift.