Back to capabilities
v0.0.3

tinysystems/embedding-module

View Source

Container Image

europe-west2-docker.pkg.dev/tinyplatform/tinysystems/embedding-module-v0:0.0.3

Components 1 available

E

embed_text

Calls a HuggingFace text-embeddings-inference (TEI) endpoint and emits the dense embedding vector. With the tei bundle enabled on install the TEI service lives in-cluster at http://-tei:80 and TEI_URL is wired automatically. Override the URL via settings.baseURL for external endpoints.

Release Notes

Tiny Systems Embedding Module

In-cluster text embeddings for Tiny Systems flows. Wraps HuggingFace text-embeddings-inference (TEI) and ships it as a curated bundle, so installing this module also provisions the TEI service in the same namespace — no external API, no separate helm install.

Components

embed_text

Takes a string, returns its dense vector embedding.

  • Input: { context: any, text: string }
  • Output: { context: any, embedding: float32[], dims: int }
  • Error: { context: any, error: string } when enabled

The component reads TEI_URL from env. The platform's install flow wires this automatically when the TEI bundle is enabled. To point at an external embedding endpoint, set baseURL in the node settings instead.

Install

The module declares the tei bundle with DefaultEnabled: true, so a plain helm upgrade --install of the operator chart pulls TEI in as a subchart and starts a Deployment + Service alongside the module.

Default model: BAAI/bge-small-en-v1.5 — 384-dim, CPU-friendly, decent quality for English RAG. Override via --set bundles.tei.image.tag=cpu-1.5 --set bundles.tei.modelId=intfloat/multilingual-e5-large if you need a different model.

The TEI Service lands at -tei:80 inside the namespace; the install flow sets TEI_URL on the module pod's env to match.

Pairs with

  • database-module vector_search / vector_upsert components (also via a bundle: pgvector) — embed + store + retrieve as a three-component RAG slice.