EmbeddingModelDefinition

One embedding model Maitento can index and search with, and the limits that decide how text is cut into chunks before it reaches it. The limits differ per model because their input ceilings differ (for example, gemini-embedding-001 accepts 2,048 tokens where text-embedding-3-large accepts 8,191).

Kind: an object, used in responses.

Fields

Every field is present in a response; empty ones are null.

Field	Type	Description
provider	EmbeddingProvider	The provider that serves the model.
model	string	The provider’s model id (for example text-embedding-3-large).
nativeDimensions	integer	The width the model produces when no output width is requested.
supportedDimensions	integer[]	The output widths a partition may be indexed at with this model.
maxInputTokens	integer	The provider’s hard input ceiling for one embedding call, in the model’s own tokens.
defaultChunkSizeTokens	integer	The chunk size used when a partition’s profile does not choose one.
defaultChunkOverlapTokens	integer	The overlap between consecutive chunks used when a partition’s profile does not choose one.
maxChunkSizeTokens	integer	The largest chunk size a profile may choose for this model.
isDefault	boolean	Whether this is the default: the model every partition without a profile of its own embeds with. Exactly one listed model is the default.
isWithdrawn	boolean	True for a model Maitento no longer offers, which a partition was indexed with before it was withdrawn. The partition can still be read, searched and re-indexed, but a new profile cannot choose it: move the partition to a listed model.
isCustom	boolean	True for your own model rather than one Maitento lists: a model id served by one of the providers’ APIs, typically an OpenAI-compatible endpoint of yours, whose width a test call measured. Its limits are the most conservative of any model, it embeds at its natural width, and it never falls back to Maitento’s key.
maxWordsPerEmbedding	integer	The most words one embedding call may carry for this model: 4,000 for an 8,191-token model, and in proportion for others. Set by Maitento.
wholeDocumentMaxWords	integer	Above this many words a document is chunked rather than embedded whole: at most 750, and never more than one call can carry. Set by Maitento.
wholeDocumentMaxCharacters	integer	Above this many characters a document is chunked rather than embedded whole: at most 30,000, and fewer for a model with a small input ceiling. Set by Maitento.

Example

{
  "provider": "OpenAi",
  "model": "...",
  "nativeDimensions": 0,
  "supportedDimensions": [
    0
  ],
  "maxInputTokens": 0,
  "defaultChunkSizeTokens": 0,
  "defaultChunkOverlapTokens": 0,
  "maxChunkSizeTokens": 0,
  "isDefault": false,
  "isWithdrawn": false,
  "isCustom": false,
  "maxWordsPerEmbedding": 0,
  "wholeDocumentMaxWords": 0,
  "wholeDocumentMaxCharacters": 0
}

Where it’s used

• partitions.embeddingModels (GET /partitions/embedding-models)