> ## Documentation Index > Fetch the complete documentation index at: https://docs.x402layer.cc/llms.txt > Use this file to discover all available pages before exploring further. # Create embeddings > OpenAI-compatible embeddings with input-only billing. Existing text models accept a string or string array. `embeddinggemma-2` also accepts ordered text, image, audio, and video content parts and returns one normalized vector per top-level item. Omitted EG2 `input_type` defaults to `query`; `unspecified` explicitly opts out. Legacy omission retains each model's existing behavior. The model is default off and is callable only when it appears in `GET /v1/models?type=embedding`. Authenticate with an `X-API-Key` carrying `grid:write` (prepaid credits), or omit it and retry the `402` challenge with `X-Payment`. EG2 input and results are persisted only as sealed envelopes. Terminal input becomes purge-eligible after 30 minutes and results after one hour; the hourly purge makes practical upper bounds about 90 minutes and two hours. Plaintext node error text is never persisted. Create normalized vectors from text or an ordered batch of text, image, audio, and video parts. EmbeddingGemma 2 is release-gated and appears in `GET /v1/models?type=embedding` only when the production flag and a capability-qualified node are ready. For payload construction, limits, Local mode, and error handling, see [Multimodal embeddings](/cloud/grid/embeddings). ## OpenAPI ````yaml POST /v1/embeddings openapi: 3.1.0 info: title: x402 Singularity Layer API description: >- OpenAPI-backed reference for marketplace discovery, payment routes, webhooks, wallet-first auth, agent endpoints, and ERC-8004 flows. version: 1.0.0 servers: - url: https://api.x402layer.cc security: [] tags: - name: Marketplace description: Public discovery and listing lookup - name: Public Endpoints description: Public endpoint metadata and hosted checkout context - name: Public Payment Links description: Hosted public payment-link lookup - name: Payments description: Hosted x402 payment challenge routes - name: Receipts description: Signed receipt lookup and verification helpers - name: Ratings description: Public listing ratings and authenticated rating actions - name: Webhooks description: Seller webhook management API - name: Agent Auth description: Wallet-first challenge and verification routes - name: Agent Endpoints description: Create, read, top up, and delete agent endpoints - name: ERC-8004 description: Agent registry and registration lifecycle routes - name: Marketplace Agents description: Public ERC-8004 marketplace discovery routes - name: Compute Catalog description: Compute plans, regions, and OS catalog - name: Compute Instances description: Provision, inspect, extend, and destroy compute instances - name: Compute API Keys description: API keys for compute agent access - name: Fundraiser Campaigns description: List, view, create, and edit fundraiser campaigns - name: Fundraiser Contributions description: Record and list campaign contributions - name: Fundraiser Comments description: Campaign comment threads - name: Fundraiser Media description: Campaign image uploads and OG images - name: Enterprise description: >- Enterprise partner configuration, endpoint listing, revenue stats, and transaction ledger - name: Staking description: Agentic $SGL staking - name: SGL Grid description: >- Decentralized, confidential, OpenAI-compatible inference served by attested TEE nodes. - name: Compute Credits description: Prepaid USDC credit balance shared across Machines and Grid. - name: Agent Pods description: >- Deploy and manage hosted agents (Agent Pods), plus the pod's OpenAI-compatible adapter for external clients. - name: Agent Pods API description: >- The programmatic `/pods/v1` surface: create, drive and destroy Agent Pods with a single `X-API-Key`. Distinct from the Agent Pods dashboard routes, which expect a wallet signature or a browser session. paths: /v1/embeddings: servers: - url: https://grid.x402compute.cc post: tags: - SGL Grid summary: Create embeddings description: >- OpenAI-compatible embeddings with input-only billing. Existing text models accept a string or string array. `embeddinggemma-2` also accepts ordered text, image, audio, and video content parts and returns one normalized vector per top-level item. Omitted EG2 `input_type` defaults to `query`; `unspecified` explicitly opts out. Legacy omission retains each model's existing behavior. The model is default off and is callable only when it appears in `GET /v1/models?type=embedding`. Authenticate with an `X-API-Key` carrying `grid:write` (prepaid credits), or omit it and retry the `402` challenge with `X-Payment`. EG2 input and results are persisted only as sealed envelopes. Terminal input becomes purge-eligible after 30 minutes and results after one hour; the hourly purge makes practical upper bounds about 90 minutes and two hours. Plaintext node error text is never persisted. parameters: - name: X-API-Key in: header required: false schema: type: string example: x402c_... description: >- Compute/Grid API key with `grid:write` scope. Requests debit prepaid credits. Omit to use the x402 challenge. - name: X-Payment in: header required: false schema: type: string description: >- x402 payment payload. Send after receiving the endpoint's `402` challenge. requestBody: required: true content: application/json: schema: type: object additionalProperties: false required: - model - input properties: model: type: string example: embeddinggemma-2 input: description: >- A single text string; 1-2048 strings for legacy text models; or 1-16 ordered string/structured items for EmbeddingGemma 2. Each item produces one vector at the matching response index. anyOf: - type: string minLength: 1 - type: array minItems: 1 maxItems: 2048 items: type: string minLength: 1 description: Legacy text-only batch. - type: array minItems: 1 maxItems: 16 items: oneOf: - type: string minLength: 1 - type: object additionalProperties: false required: - content properties: content: type: array minItems: 1 maxItems: 16 description: >- Ordered content parts. Up to eight images, one audio part, and one video part per item. items: oneOf: - type: object title: Text part additionalProperties: false required: - type - text properties: type: type: string enum: - text text: type: string minLength: 1 - type: object title: Image part additionalProperties: false required: - type - media properties: type: type: string enum: - image media: type: object additionalProperties: false required: - encoding - mime_type - data - sha256 properties: encoding: type: string enum: - base64 mime_type: type: string enum: - image/jpeg - image/png - image/webp data: type: string description: >- Canonical base64 for at most 8 MiB decoded bytes. sha256: type: string pattern: ^[a-f0-9]{64}$ description: >- SHA-256 of the decoded bytes, lowercase hexadecimal. - type: object title: Audio part additionalProperties: false required: - type - media - duration_seconds properties: type: type: string enum: - audio duration_seconds: type: number exclusiveMinimum: 0 maximum: 30 description: >- Declared duration. The runtime verifies the decoded duration. media: type: object additionalProperties: false required: - encoding - mime_type - data - sha256 properties: encoding: type: string enum: - base64 mime_type: type: string enum: - audio/wav - audio/flac - audio/mpeg data: type: string description: >- Canonical base64 for at most 8 MiB decoded bytes. sha256: type: string pattern: ^[a-f0-9]{64}$ - type: object title: Video part additionalProperties: false required: - type - media - duration_seconds properties: type: type: string enum: - video duration_seconds: type: number exclusiveMinimum: 0 maximum: 32 description: >- Declared duration. Video is sampled at up to 1 fps and 32 frames. media: type: object additionalProperties: false required: - encoding - mime_type - data - sha256 properties: encoding: type: string enum: - base64 mime_type: type: string enum: - video/mp4 data: type: string description: >- Canonical base64 for at most 16 MiB decoded bytes. sha256: type: string pattern: ^[a-f0-9]{64}$ dimensions: description: >- Optional model-specific output dimension. Omit for the model's native dimension. Unsupported model/dimension pairs return 400. anyOf: - title: EmbeddingGemma 2 type: integer enum: - 768 - 512 - 256 - 128 - title: Nomic Embed Text v1.5 type: integer enum: - 768 - 512 - 256 - 128 - 64 - title: Mixedbread Embed Large v1 type: integer enum: - 1024 - 512 - 256 - title: Fixed-size text models type: integer enum: - 384 - 768 - 1024 description: >- The selected fixed-size model accepts only its own native value. input_type: description: >- Optional model-specific retrieval behavior. This schema intentionally has no global default. anyOf: - title: Legacy text embedding models type: string enum: - query - document description: Omission retains the legacy model/runtime behavior. - title: EmbeddingGemma 2 type: string enum: - query - document - unspecified description: >- Omission is sealed as query. Use unspecified only to opt out of query/document prefixing. encoding_format: type: string enum: - float default: float tier: type: string enum: - standard - confidential use_credits: type: boolean description: >- Use an authenticated wallet session's credits when no API key is supplied. user: type: string description: Optional OpenAI-compatible caller identifier. examples: text: summary: Text batch value: model: embeddinggemma-2 input: - search query - document text dimensions: 256 input_type: query mixed: summary: Ordered text and image value: model: embeddinggemma-2 input: - content: - type: text text: Product photo - type: image media: encoding: base64 mime_type: image/png data: iVBORw0KGgo... sha256: >- 0123456789abcdef0123456789abcdef0123456789abcdef0123456789abcdef dimensions: 256 input_type: document responses: '200': description: >- One finite, unit-normalized float vector per input item, ordered by index. content: application/json: schema: type: object required: - object - data - model - usage properties: object: type: string enum: - list data: type: array items: type: object required: - object - index - embedding properties: object: type: string enum: - embedding index: type: integer minimum: 0 embedding: type: array items: type: number format: float model: type: string usage: type: object required: - prompt_tokens - total_tokens - cost_usd properties: prompt_tokens: type: integer minimum: 1 total_tokens: type: integer minimum: 1 description: >- Equal to prompt_tokens. Embeddings have no output tokens. cost_usd: type: number minimum: 0 breakdown: type: object description: Present for EmbeddingGemma 2. required: - text - image - audio - video properties: text: type: integer minimum: 0 image: type: integer minimum: 0 audio: type: integer minimum: 0 video: type: integer minimum: 0 processor_revision: type: string description: Present for EmbeddingGemma 2. embedding_protocol: type: string enum: - embedding-multimodal-v1 billing_pending: type: boolean description: >- Present and true only when x402 capture succeeded but durable accounting finalization is still recovering. Do not resubmit the paid request. job_id: type: string format: uuid description: >- Present with billing_pending so accounting recovery is correlated idempotently. '400': description: >- `invalid_request_error`; stable EG2 codes are `embedding_input_invalid` and `embedding_context_overflow`. These input failures are not charged and do not trigger paid failover. '401': description: '`invalid_api_key` or `invalid_session`.' '402': description: >- `payment_required`, `insufficient_credits`, `pod_cap_reached`, or `payment_error`. '403': description: '`insufficient_scope`: the API key lacks `grid:write`.' '404': description: >- `model_not_found` for an unknown model id. When the entire embeddings route is disabled, the route returns plain `Not found`. '413': description: Encoded JSON body exceeds 24 MiB. '500': description: >- `server_error`: dispatch or billing infrastructure failed. Inspect the response message before retrying. '502': description: >- `inference_error`: unreadable, invalid, or failed node output. Invalid vectors are never returned or billed. '503': description: >- `model_not_available` when the release/capability boundary is unavailable; `node_not_available` when a claimed node loses negotiated confidential transport; or `server_error` when the payment service is not configured. '504': description: '`timeout`: inference timed out and was not charged.' ```` This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.