Extract Text Embedding (Ollama)

How it fits

Typical backends

Caveats

• Hard constraintA reachable Ollama runtime must be deployed alongside this component; without it the embedding calls fail. The model named by model_name is pulled into that runtime on first use, adding a one-time delay.
• Hard constraintThe embedding dimension is fixed by the chosen model_name; switching models produces vectors of a different width that an existing index will reject.
• I/O contractAn empty String short-circuits to an empty Embedding. Feeding that into a similarity stage breaks the math, so filter empty inputs upstream.
• I/O contractA document longer than the model context is truncated to fit and still returns one Embedding; it is never rejected. Chunk long documents upstream when the trailing text matters, since only the leading portion is embedded.
• Parameter interactionnormalize produces unit-length vectors so cosine similarity equals the inner product; keep it consistent across the index and the queries.
• LatencyThroughput depends on the external runtime and the model size; large models on shared hardware raise per-call latency.
• State lifetimeThe model selection and normalisation are read once at startup; changing them requires a redeploy.