# Credentials (https://docs.neuralisapp.com/brain-core/credentials)

brain-core declares four credentials, one per keyed cloud embedding provider.
They are separate from the chat model keys and kept in the
[credential store](https://docs.neuralisapp.com/package-system/credentials.md) — packages never read
embedding secrets from the process environment.

| Credential id | Provider |
| --- | --- |
| `embedding.openai` | OpenAI |
| `embedding.gemini` | Gemini |
| `embedding.qwen` | Qwen / DashScope |
| `embedding.voyage` | Voyage AI |

A keyed [embedding endpoint](https://docs.neuralisapp.com/brain-core/config-settings.md#embedding) — any
server that speaks the OpenAI `/v1/embeddings` protocol — keeps its key as the
credential `embedding.endpoint.<id>`, never in the endpoint entry. Ollama, the
local default, needs no key.

These keys are read at the **platform (global) scope only**. Indexing is
background work with no user in context, and one collection serves every
project, so a member's own value for the same id is never consulted. Because
they are not chat-model keys, the per-credential use limits count them, and
embedding spend is metered in dollars per call against `embeddingDailyUsdCap`.
A keyed provider is never called without its credential: a missing one raises
`EMBEDDING_CREDENTIAL_MISSING` naming the id and its scope, content keeps being
indexed, and semantic search falls back to text search with a warning (see
[Sources](https://docs.neuralisapp.com/brain-core/sources.md#the-embedding-pipeline)). Saving a credential
alone does not start a paid rebuild.

The vector database's own API key is not a credential-store entry: it is part
of the deployment's environment, minted by setup (see
[configuration](https://docs.neuralisapp.com/enterprise/configuration.md)).
