Feature × Audience

How do developers build embeddings search on Plugsky?

For developers, the win is a familiar endpoint and a disciplined pipeline: version the model, respect the dimension, and evaluate on real queries. Plugsky exposes an OpenAI-compatible /v1/embeddings endpoint with the plugsky-embed family in a 30+ model catalogue, 8K-class inputs and a vector dimension published per model that you read before creating the collection.

Key facts

EndpointPOST /v1/embeddings (OpenAI-compatible)
Modelsplugsky-embed, plugsky-embed-multilingual and plugsky-embed-nim in a 30+ model catalogue
Input limit8K-class text per request; live limits published per model
Vector dimensionPublished per model on /models — read it before creating a collection
Use casesRAG, semantic search, clustering and recommendations
IndexingBatch embeddings for large corpora; re-embed when the model version changes
API compatibilityOpenAI-compatible /v1/embeddings; change base URL, keep your code
Pipeline designChunk by structure; read dimension from /models before indexing

TL;DR

  • OpenAI-compatible /v1/embeddings with the plugsky-embed family in a 30+ model catalogue.
  • Dimension is published per model — read it before creating the collection.
  • One OpenAI-compatible endpoint for chat and embeddings keeps integration simple.
  • Dimension is a schema contract; record it with every vector.
  • Start free with plugsky-micro and plugsky-lite; a 14-day full-access trial covers larger models.

How it works, step by step

  1. Read the model's live vector dimension from the catalogue and create the collection with it.
  2. Chunk by document structure, embed in batches, and store metadata for citations.
  3. Validate retrieval on a labelled question set before moving to production traffic.
  4. Read the model's live dimension and create the collection with it.
  5. Batch-index with structure-aware chunks and citation metadata.
  6. Evaluate retrieval on your own queries before adding a reranker.
1Read the model'slive vectordimension from the2Chunk by documentstructure, embed inbatches, and store3Validate retrievalon a labelledquestion set before4Read the model'slive dimension andcreate the5Batch-index withstructure-awarechunks and citation6Evaluate retrievalon your own queriesbefore adding a

Original data

POST /v1/embedEndpointplugsky-embed,Models8K-class text Input limitOpenAI-compatiAPI compatibilitySource: Plugsky facts table · updated 2026-09-26

Try it yourself

Open the embedding API tester →

Embeddings for developers: what changes

Developers adopt embeddings to power search, recommendations, deduplication and RAG. The work is familiar: call an endpoint, store vectors, query by similarity. What matters is picking a model, respecting the dimension contract, and keeping the pipeline maintainable when models change.

Embeddings run on an OpenAI-compatible /v1/embeddings endpoint, so existing vector pipelines keep their request shape. The plugsky-embed family — plugsky-embed, plugsky-embed-multilingual and plugsky-embed-nim — sits alongside chat models in a 30+ model catalogue, each with an 8K-class input limit and a published vector dimension you can read from the catalogue before indexing.

Architecture and controls

Call /v1/embeddings with the same authentication and workspace model used for chat, batch indexing jobs, and read the live vector dimension from the model catalogue before creating a collection. Validate inputs against the model's input limit.

Integration pattern and rollout

Build the smallest useful loop first: chunk documents, embed, store with metadata, retrieve top-k, then add hybrid keyword search and a reranker if precision needs it. Keep model choice in configuration so switching is a config diff.

The pipeline is straightforward: chunk, embed in batches, store vectors with source metadata, retrieve top-k. What needs discipline is versioning — record the model and dimension with every vector, keep model choice in configuration, and treat a model swap as a data migration with dual-write and a validation phase before cutover.

Limits, evidence and cost

Switching embedding models invalidates stored vectors; budget for dual-write and backfill rather than an in-place migration. Quality varies by corpus and language, so evaluate on your own data instead of relying on generic comparisons.

Self-serve plans are flat monthly with unlimited fair-use usage, so embedding volume does not introduce per-token billing — check the live pricing page. The free plan includes plugsky-micro and plugsky-lite with no card, and the 14-day full-access trial lets you test before committing to an index.

Honest comparison

ConcernPlugsky embeddingsTypical API providerSelf-hosted embedder
API shapeOpenAI-compatible /v1/embeddingsUsually compatible, variesCustom serving stack
Model choiceplugsky-embed family inside a 30+ model catalogueProvider catalogue onlyYou package each model
ResidencyRegion-locked planes; VPC, on-prem and air-gappedLimited region choicesWherever you deploy
Dimension changesRead live dimension from /models; plan new collectionsVaries by providerYou manage every migration
Operational loadManaged endpoint with batchingManaged endpointGPU capacity, patching and autoscaling
API shapeOpenAI-compatible /v1/embeddingsUsually compatibleCustom serving layer

Frequently asked questions

Do we need to re-embed when we change models?

Yes. Vectors depend on the model and dimensions can change. Plan a new collection, backfill, validate, then cut over — re-embedding is a data migration, not a config flip.

Are embeddings region-locked?

Yes, when the workload is pinned to a region-locked plane. Keep the vector store, backups and logs in the same jurisdiction as the source data.

Which embedding model should we start with?

plugsky-embed for English-dominant corpora, plugsky-embed-multilingual for mixed languages, and plugsky-embed-nim for NVIDIA-style profiles. Evaluate on your own content.

Can we keep our OpenAI embedding code?

Yes. Point the base URL at Plugsky and keep the request shape; read the live dimension from the model catalogue before creating the collection.

How do we handle long documents?

Chunk by structure to fit the 8K-class input limit, keep metadata for citations, and batch indexing jobs to control load.

Should we add a reranker?

If top-k precision matters, combine vector search with keyword matching and a reranker. Evaluate on your own queries rather than assuming a gain.