Pricing
Retrieval is included. You pay for storage.
No separate plan, no separate quota.
What’s billed
Three lines, and only one of them is ours.
Retrieval
Every retrieval, rerank and query. No additional charge, no per-query counter.
$0Storage
$0.10 per GB of indexed data, per day. This is the only GTWY line item on this product.
$0.10 / GB / dayEmbedding
You bring your own embedding key. Embedding costs go straight to your provider — we never mark them up, because we never touch them.
your rateUsage
- Unlimited retrieval inside agent runs
- Full ingest pipeline — parse, chunk, embed, index
- Every connected source type
- The embeddable upload UI
- Per-customer indexes and storage caps
Enterprise
- Dedicated infrastructure
- Custom terms, SLA and DPA
- Raised ingest and storage limits
- Security review support
- Data residency options
Calculator
What storage costs you, and what you can charge.
Retrieval doesn’t appear here because it isn’t billed. Storage is the whole GTWY line.
Rate applied: $0.10 / GB / day · totals shown per 30-day month
75% of what you bill is yours. Embedding costs sit on your own provider account and are not included above.
Bring your own key
What a re-index costs you, and when it happens.
Under BYOK a re-index bills against your provider account, not ours. That removes any markup question — and makes it our job to make sure you see it coming.
What a re-index does
Re-embeds every document for every affected customer. On a large platform that is a real number of tokens, and it lands on your provider bill.
What triggers one
Changing the embedding model, changing chunking configuration, or a migration. Nothing else re-embeds content that is already indexed.
You see the cost first
Estimated token volume and cost are shown before you confirm. A re-index never starts without that preview.
If your key stops working
An invalid, revoked or rate-limited key halts ingest rather than failing silently. Documents queue, the panel and your alerting show the failure, and ingest resumes when the key does — so nobody discovers it through an agent that quietly stopped knowing things.
Questions
Before you model this out.
Is retrieval really not charged per query?
Really. Retrieval, reranking and every knowledge lookup happen inside the agent run you are already paying for. There is no per-query meter to watch.
What counts toward storage — original files, chunks, or embeddings?
Indexed data: the extracted text and its vectors. Original files are retained so documents can be re-processed and exported, and are not counted twice.
Can we set per-customer storage caps?
Yes. A default cap applies to every customer and you override it per customer, so one customer uploading their entire document history doesn’t become your bill.
What happens if a customer exceeds their storage limit?
New uploads are refused at the cap; everything already indexed stays searchable. Your user sees the limit in the upload UI, and you get the signal to raise it or sell an upgrade.
Whose embedding key is used — ours or our customers’?
Pending decision. Whether the key is one per platform or one per end customer changes whether you can attribute embedding spend per customer. Not published until it’s settled.
What happens if our embedding key stops working?
Ingest halts and alerts rather than failing quietly. Queued documents resume once the key is valid again; nothing already indexed is affected, so retrieval keeps working on what is already there.
Will changing the embedding model cost us a full re-embed?
Yes — the index is built to a model’s dimensionality, so switching re-embeds everything. You see the estimated token cost before confirming, and it bills to your provider account.