Documentation
¶
Overview ¶
Package embedmodel bundles a sentence-embedding model so agentic memory works with zero external dependencies — no API key, no network call (after the one-time model download), no separate model server required.
The model is thenlper/gte-small (384-dimensional embeddings), run via Hugot's pure-Go (GoMLX simplego) backend. Because Hugot loads models from filesystem paths, the small tokenizer/config assets are embedded in the binary and lazily materialized to a cache directory on first use, while the large ONNX weight file (~133 MB) is downloaded on first use from Hugging Face rather than bloating the binary, which keeps the distributable binary small while preserving the single-command, no-API-key experience.
gte-small was chosen over the previous all-MiniLM-L6-v2 because it scores higher on the MTEB benchmark (~61.4 vs ~56.3), requires no query/passage prefixing (unlike bge-small or e5-small), and is a drop-in 384-dim replacement.
Index ¶
Constants ¶
const EmbeddingDim = 384
EmbeddingDim is the dimensionality of the vectors produced by the model.
const ModelFileName = "model.onnx"
ModelFileName is the name of the ONNX weight file in the cache directory.
const ModelName = "thenlper/gte-small"
ModelName is the human-readable identifier of the bundled model.
const ModelSHA256 = "0b01312b59bec0a2558a626f2937be4cbe4bb16d1511560153f598cec488f1f8"
ModelSHA256 is the expected SHA-256 of the downloaded ONNX file, used to verify integrity and detect partial/corrupt downloads.
const ModelURL = "https://huggingface.co/thenlper/gte-small/resolve/main/onnx/model.onnx"
ModelURL is the canonical download URL for the ONNX weights. The file is fetched on first use when it is not already present in the cache directory.
Variables ¶
var AssetNames = []string{
"tokenizer.json",
"tokenizer_config.json",
"special_tokens_map.json",
"config.json",
"vocab.txt",
}
AssetNames is the ordered list of small files bundled in the binary, relative to the assets/ directory. Each is written verbatim to the cache dir. The large ONNX model file is NOT in this list — it is downloaded on first use (see ModelURL / ModelSHA256).
Functions ¶
Types ¶
This section is empty.
Source Files