Costs and Budgets
What a knowledge base spends on indexing and searching, who pays for it, the price of a rebuild before it starts, a base's costs this month, and a monthly budget that pauses instead of overspending.
A knowledge base spends money in a few places, and you can see each of them before and after:
| Work | What it spends on |
|---|---|
| Indexing a document | Embedding every passage. With Contextual Retrieval, a model call per passage too. On SynapseOrch Cloud, reading scanned pages by OCR |
| Rebuilding the index | Embedding again — and contexts again, if the base writes them. Files are not read again |
| A search | Embedding the question, and the rerank when one is set |
| Evaluate | The searches it runs, and the model calls that draft questions and grade answers |
Reading files is free: it runs on the server.
Who pays
On a self-hosted install your providers bill you directly; Synapse estimates each cost from Usage's rate card. A local model costs nothing, and a model with no price on the rate card counts as $0 — it is never stopped by a budget, and its spend is not in the totals. See Usage logs.
On SynapseOrch Cloud SynapseOrch's own models are charged to your model wallet, at the price below. The monthly allowance pays for them, as it does for any model. A model on your own provider key costs the wallet nothing: the provider bills you, and Synapse shows the estimate.
Prices on SynapseOrch Cloud
| Charged | |
|---|---|
| Embedding with Titan Text Embeddings v2 (the default) | $0.025 per million tokens |
| Reranking with SynapseOrch's reranker (the default while it answers) | $0.0025 per search — one question against up to 100 passages |
| Reading a scanned page with Textract | $0.001875 per page read; pages with text are never sent |
| Contextual Retrieval with Nova Lite (its default) | $0.075 per million input tokens, $0.30 per million output |
| A workflow's Retrieve or Ingest step | 0.1 credit, as a step; the search or the indexing it starts is charged as above |
| Storage | A base's files count toward your plan's Vault storage; the index and what is cached to rebuild it do not |
Each price is the provider's list price with the markup every plan carries. There is no fee per search and no charge for the index itself. Embedding counts each passage together with its title-and-section line, so a document embeds at roughly one and a half times its own token count.
The price before you spend
Whenever a change to a base's settings would rebuild its index, the settings show what the rebuild will cost before you save:
Saving rebuilds the index: 24 documents, about $0.0042. The current index keeps answering until the new one is ready. Embedding: 168,000 tokens with synapse.titan-embed-v2 — $0.0042
A line ends (your key) when a model on your own key will bill you instead of the wallet, and (on this machine) for a local model. If the price is more than the budget has left this month, it says so: the rebuild will pause at the budget. The Knowledge API's cost-preview prices any set of settings the same way, without saving them.
Costs this month
The Costs section on a base's page shows what it has spent this month, by stage — Embedding, Reranking, Reading scanned pages, Context — or by day, by document or by model. Searches appear as one line, Searches. On SynapseOrch Cloud each line shows what the wallet was charged beside the estimate.
A base's spend also reaches the rest of Synapse:
- A Retrieve step's search is part of its run's cost.
- Usage lists Knowledge's own spend as one session per base:
knowledge:and the base's id on a self-hosted install, the base's name on SynapseOrch Cloud. - On SynapseOrch Cloud, Billing → Model spend shows each charge — N search units for a rerank, N pages for OCR.
Budgets
Monthly budget (USD), in a base's settings, caps what it spends in a calendar month (UTC). Synapse checks the estimate before each batch of work, not after:
- Indexing pauses at the budget. Documents waiting are Paused with Monthly budget reached; a rebuild pauses with Paused: the monthly budget is reached. Raise it to go on, or it resumes next month. Nothing is retried in a loop, and nothing already done is paid for again when it goes on.
- Searches carry on with words only. A base at its budget is still searched — by keyword, with no question embedding and no rerank charged — and the search says This knowledge base hit its monthly budget.
- It goes on by itself when you raise or clear the budget, or when the month turns.
The Costs section shows the budget as a meter: $3.20 left of a $5.00 monthly budget.
In Evaluate, the searches count toward the budget, and the model calls that draft questions or grade answers do not; a run that would cross the budget is refused before it starts.
An empty wallet (SynapseOrch Cloud)
When the model wallet can't pay, indexing pauses the same way — Paused, Model wallet can't pay — and goes on by itself within the hour of a top-up, paying only for what it had not done: passages not yet embedded, contexts not yet written, pages not yet read.
Search and Settings
How a knowledge base search finds passages — words, meaning, fusion and rerank — the test search and its trace, every setting, embedding and rerank models, and rebuilds.
Contextual Retrieval
Have a model write a short context for every passage, so passages that lean on their document are found — what it costs, when it helps, and what we measured.