Embeddings

Show: All versions v2 only

Schift provides a unified embedding API that proxies to multiple providers (OpenAI, Google, Voyage, Cloudflare, HuggingFace, and more) through a single set of endpoints. All embeddings pass through Schift’s Canonical Projection layer, so you can switch models or reduce dimensions without re-embedding your data.

Authentication

All embedding endpoints require a Bearer token in the Authorization header:

Authorization: Bearer sch_xxxxxxxxxxxxxxxxxxxx

API keys are managed in the Schift dashboard under Settings > API Keys.

Supported models

Model	Provider	Default dimension	Max tokens	Variable dimensions
schift-embed-1-small	Schift (auto)	1024	8,192	Yes
openai/text-embedding-3-small	OpenAI	1536	8,191	Yes
openai/text-embedding-3-large	OpenAI	3072	8,191	Yes
google/gemini-embedding-001	Google	3072	8,191	Yes
google/gemini-embedding-002	Google	3072	8,191	Yes
voyage/voyage-4-large	Voyage AI	1024	32,000	Yes
voyage/voyage-4	Voyage AI	1024	32,000	Yes
voyage/voyage-4-lite	Voyage AI	1024	32,000	Yes
dragonkue/bge-m3-ko	HuggingFace	1024	8,192	No
jinaai/jina-embeddings-v3	HuggingFace	1024	8,194	Yes
sbintuitions/sarashina-embedding-v2-1b	HuggingFace	1792	8,192	No

schift-embed-1-small is an auto-routed alias. Schift selects the best underlying model based on the input language and passes the result through the canonical projection layer. The resolved model is returned in the model field of every response.

Note: Internal backend aliases such as @cf/qwen/qwen3-embedding-0.6b may be returned as the resolved model when schift-embed-1-small is used.

Task types

Embedding endpoints optionally accept a task_type that tells Schift how the embedding will be used. The value is passed through to instruct-aware models or converted into an internal prefix for prefix-style models.

task_type	Use case
retrieval_query	Search query embedding
retrieval_document	Document embedding
semantic_similarity	Semantic similarity comparison
question_answering	Question-answering retrieval
clustering	Clustering or topic grouping
classification	Classification
code_retrieval	Code search
contradiction	Contradiction or counter-evidence search
factcheck	Fact-checking evidence search

POST /v1/embed

Embed a single text string.

Request body

Parameter	Type	Required	Default	Description
text	string	Yes	—	The text to embed.
model	string	No	Org routing config	Model ID from the catalog.
dimensions	integer	No	Model default	Output dimension. Only supported by models with variable dimensions.
task_type	string	No	—	Embedding intent. See Task types.

Response fields

Field	Type	Description
embedding	number[]	The embedding vector.
model	string	The model that was actually used.
dimensions	integer	The output dimension of the embedding.
usage.tokens	integer	Number of tokens consumed.

Example request

curl https://api.schift.io/v1/embed \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $SCHIFT_API_KEY" \
  -d '{
    "text": "Schift enables seamless embedding model migration without re-embedding your data.",
    "model": "openai/text-embedding-3-large"
  }'

Example response

{
  "embedding": [-0.01225, 0.00207, 0.03060],
  "model": "openai/text-embedding-3-large",
  "dimensions": 3072,
  "usage": {"tokens": 14}
}

POST /v1/embed/batch

Synchronous batch embedding. Processes up to 100 texts in a single request. For larger inputs, use POST /v1/embed/jobs.

Request body

Parameter	Type	Required	Default	Description
texts	string[]	Yes	—	List of texts to embed (max 100).
model	string	No	Org routing config	Model ID from the catalog.
dimensions	integer	No	Model default	Output dimension.
task_type	string	No	—	Embedding intent.

Response fields

Field	Type	Description
embeddings	number[][]	List of embedding vectors, one per input text.
model	string	The model that was used.
dimensions	integer	The output dimension.
usage.tokens	integer	Total tokens across all texts.
usage.count	integer	Number of texts embedded.

Example request

curl https://api.schift.io/v1/embed/batch \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $SCHIFT_API_KEY" \
  -d '{
    "texts": [
      "The Mediterranean diet emphasizes fish, olive oil, and vegetables.",
      "Photosynthesis converts light energy into chemical energy.",
      "Shakespeare wrote Hamlet and A Midsummer Night'"'"'s Dream."
    ],
    "model": "voyage/voyage-4-large"
  }'

Example response

{
  "embeddings": [
    [-0.01225, 0.00207, 0.03060],
    [0.00898, -0.00298, 0.01546],
    [-0.06297, -0.04777, -0.10113]
  ],
  "model": "voyage/voyage-4-large",
  "dimensions": 1024,
  "usage": {"tokens": 38, "count": 3}
}

POST /v1/embed/jobs

Create an asynchronous bulk embedding job. Use this endpoint for 101 to 10,000 texts.

Request body

Parameter	Type	Required	Default	Description
texts	string[]	Yes	—	List of texts to embed (max 10,000).
model	string	No	Org routing config	Model ID from the catalog.
dimensions	integer	No	Model default	Output dimension.
task_type	string	No	—	Embedding intent.

Response fields

Field	Type	Description
id	string	Bulk embed job ID.
status	string	Initial status (queued).
model	string	Model selected for the job.
usage.tokens	integer	Total validated tokens across all texts.
usage.count	integer	Number of texts submitted.
chunk_size	integer	Worker chunk size (currently 100).

Example request

curl https://api.schift.io/v1/embed/jobs \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $SCHIFT_API_KEY" \
  -d '{
    "texts": ["doc 1", "doc 2", "doc 3"],
    "model": "schift-embed-1-small",
    "task_type": "retrieval_document"
  }'

Example response

{
  "id": "job_123",
  "status": "queued",
  "model": "schift-embed-1-small",
  "usage": {"tokens": 6, "count": 3},
  "chunk_size": 100
}

GET /v1/embed/jobs/{job_id}

Retrieve the status and metadata of a bulk embed job.

Path parameters

Parameter	Type	Description
job_id	string	The bulk embed job ID.

Job status lifecycle

Status	Meaning
queued	Job created and waiting for a worker.
embedding	Worker is embedding text chunks.
indexing	Results are being prepared and stored.
ready	Results are available for retrieval.
failed	Job failed.
cancelled	Job was cancelled before processing.

Example request

curl https://api.schift.io/v1/embed/jobs/job_123 \
  -H "Authorization: Bearer $SCHIFT_API_KEY"

Example response

{
  "id": "job_123",
  "status": "ready",
  "org_id": "org_abc",
  "metadata": {
    "mode": "embed_bulk",
    "model": "schift-embed-1-small",
    "dimensions": null,
    "task_type": "retrieval_document",
    "input_count": 3,
    "chunk_size": 100,
    "token_count": 6,
    "completed_count": 3
  }
}

POST /v1/embed/jobs/{job_id}/cancel

Cancel a queued bulk embed job. Only jobs in queued status can be cancelled.

Path parameters

Parameter	Type	Description
job_id	string	The bulk embed job ID.

Example request

curl -X POST https://api.schift.io/v1/embed/jobs/job_123/cancel \
  -H "Authorization: Bearer $SCHIFT_API_KEY"

Example response

{
  "status": "cancelled"
}

GET /v1/embed/jobs/{job_id}/result

Retrieve the paginated results of a completed bulk embed job.

Path parameters

Parameter	Type	Description
job_id	string	The bulk embed job ID.

Query parameters

Parameter	Type	Default	Description
offset	integer	0	Number of results to skip.
limit	integer	100	Maximum number of results to return (max 1,000).

Response fields

Field	Type	Description
object	string	Always list.
data	object[]	Paginated embedding result items.
model	string	Model used for the result.
dimensions	integer	Output dimension.
usage.count	integer	Total embedded item count.
pagination.offset	integer	Requested offset.
pagination.limit	integer	Requested limit.
pagination.returned	integer	Number of items returned in this page.
pagination.total	integer	Total result item count.
pagination.has_more	boolean	Whether more pages remain.

Example request

curl "https://api.schift.io/v1/embed/jobs/job_123/result?offset=0&limit=100" \
  -H "Authorization: Bearer $SCHIFT_API_KEY"

Example response

{
  "object": "list",
  "data": [
    {"object": "embedding", "index": 0, "embedding": [-0.01225, 0.00207]},
    {"object": "embedding", "index": 1, "embedding": [0.00898, -0.00298]},
    {"object": "embedding", "index": 2, "embedding": [-0.06297, -0.04777]}
  ],
  "model": "schift-embed-1-small",
  "dimensions": 1024,
  "usage": {"count": 3},
  "pagination": {
    "offset": 0,
    "limit": 100,
    "returned": 3,
    "total": 3,
    "has_more": false
  }
}

POST /v1/embed/image

Embed images by first extracting text with a vision-language model, then embedding the extracted text. Accepts up to 20 base64-encoded images per request.

Request body

Parameter	Type	Required	Default	Description
images	string[]	Yes	—	Base64-encoded image data.
model	string	No	Org routing config	Model ID from the catalog.
dimensions	integer	No	Model default	Output dimension.

Response fields

Field	Type	Description
embeddings	number[][]	List of embedding vectors, one per image.
model	string	The model that was used.
dimensions	integer	The output dimension.
usage.image_count	integer	Number of images processed.
usage.tokens	integer	Total tokens consumed.

Example request

curl https://api.schift.io/v1/embed/image \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $SCHIFT_API_KEY" \
  -d '{
    "images": ["iVBORw0KGgoAAAANSUhEUg..."],
    "model": "openai/text-embedding-3-large"
  }'

Example response

{
  "embeddings": [[-0.01225, 0.00207, 0.03060]],
  "model": "openai/text-embedding-3-large",
  "dimensions": 3072,
  "usage": {"image_count": 1, "tokens": 14}
}

POST /v1/embeddings

OpenAI-compatible embeddings endpoint. Accepts the same input shape as OpenAI’s /v1/embeddings and returns an OpenAI-compatible response shape. This endpoint is a thin wrapper over the same logic as /v1/embed and /v1/embed/batch.

Note: For more than 100 inputs, use POST /v1/embed/jobs instead.

Request body

Parameter	Type	Required	Default	Description
input	string | string[]	Yes	—	Text or list of texts to embed.
model	string	No	Org routing config	Model ID from the catalog.
dimensions	integer	No	Model default	Output dimension.
task_type	string	No	—	Schift extension: embedding intent.
encoding_format	string	No	—	Accepted for compatibility, currently unused.
user	string	No	—	Accepted for compatibility, currently unused.

Response fields

Field	Type	Description
object	string	Always list.
data	object[]	Embedding objects with object, index, and embedding.
model	string	The model that was used.
usage.prompt_tokens	integer	Total tokens consumed.
usage.total_tokens	integer	Total tokens consumed.

Example request

curl https://api.schift.io/v1/embeddings \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $SCHIFT_API_KEY" \
  -d '{
    "input": "Schift enables seamless embedding model migration.",
    "model": "openai/text-embedding-3-large"
  }'

Example response

{
  "object": "list",
  "data": [
    {
      "object": "embedding",
      "index": 0,
      "embedding": [-0.01225, 0.00207, 0.03060]
    }
  ],
  "model": "openai/text-embedding-3-large",
  "usage": {
    "prompt_tokens": 8,
    "total_tokens": 8
  }
}

Canonical projection

Every embedding returned by Schift is projected into a shared canonical latent space. This enables:

• Model switching without re-embedding. Move from OpenAI to Voyage without touching your vector store.
• Cross-model search. Query vectors from one model can retrieve documents embedded with another.
• Dimension reduction. Request any output dimension regardless of the source model’s native dimension.

The projection happens transparently. Call the API normally and Schift handles the rest.

Routing and failover

Your organization can configure a default embedding model and optional fallback. When failover mode is enabled, Schift automatically retries with the fallback model if the primary provider is unavailable.

Use the routing endpoints to configure defaults:

curl -X PUT https://api.schift.io/v1/routing \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $SCHIFT_API_KEY" \
  -d '{
    "primary": "openai/text-embedding-3-large",
    "fallback": "voyage/voyage-4-large",
    "mode": "fallback"
  }'

The response from every embed endpoint includes the model field so you always know which model was used.

Rate limits and quotas

Rate limits are enforced per organization. Usage is tracked in real time. The following axes are checked for embedding endpoints:

Endpoint	Quota axis
POST /v1/embed	embed
POST /v1/embed/batch	embed_batch
POST /v1/embed/image	embed_image
POST /v1/embed/jobs	embed_batch
POST /v1/embeddings	embed or embed_batch

Retrieve current usage with:

curl https://api.schift.io/v1/usage/summary \
  -H "Authorization: Bearer $SCHIFT_API_KEY"

Bulk embed jobs are assigned a priority based on your organization’s plan:

Plan	Priority
Enterprise	0
Business	1
Pro / Paid / Starter	2
Free	3

Lower numbers are processed first.

Error codes

Common errors

Status	Meaning
400	Bad request — unknown model, invalid dimensions, empty input, or token limit exceeded.
401	Invalid or expired API key.
402	Insufficient credits or quota exceeded.
403	Feature not available on your plan or entitlement limit reached.
502	Upstream embedding provider failed (both primary and fallback).

Bulk job errors

Endpoint	Status	Meaning
POST /v1/embed/batch	400	More than 100 texts in a sync request. Use POST /v1/embed/jobs.
POST /v1/embed/jobs	400	Empty texts, more than 10,000 texts, or a text exceeds the model’s max_tokens.
GET /v1/embed/jobs/\{job_id\}	404	Job not found or belongs to another organization.
POST /v1/embed/jobs/\{job_id\}/cancel	400	Job is already processing or completed.
POST /v1/embed/jobs/\{job_id\}/cancel	409	Job is no longer queued.
GET /v1/embed/jobs/\{job_id\}/result	404	Job or result not found.
GET /v1/embed/jobs/\{job_id\}/result	409	Result not ready yet.

See also

• Buckets
• Search
• Usage
• Routing