Repository navigation
Conversation
The spec explicitly allows memoryfields to use embedding models other than nomic-embed-text-v1.5, but the tool hardcodes it. This makes the embedding model configurable through the environment: - MEMORYFIELD_EMBED_MODEL: any model available in the local ollama (eg: bge-m3, a strong multilingual model that handles paraphrase queries much better than nomic-embed-text in non-English languages) - MEMORYFIELD_EMBED_MODEL_CODE: optional explicit model code for the vector index filename (defaults to the ollama model name, per the spec's note on provider shorthands) With no env vars set, behaviour is unchanged. The index filename follows MODEL_CODE, so different models produce separate index caches and old indexes remain valid.
This branch has not been deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The spec explicitly allows memoryfields to use embedding models other than nomic-embed-text-v1.5, but the tool hardcodes it. This makes the embedding model configurable through the environment:
MEMORYFIELD_EMBED_MODEL: any model available in the local ollama (eg:bge-m3, a strong multilingual model that handles paraphrase queries much better than nomic-embed-text in non-English languages — in our testing with 14 Russian paraphrase queries, bge-m3 ranked the correct page first 11/14 times vs 2/14 for nomic-embed-text)MEMORYFIELD_EMBED_MODEL_CODE: optional explicit model code for the vector index filename (defaults to the ollama model name, per the spec's note on provider shorthands)With no env vars set, behaviour is unchanged (covered by a test). The index filename follows MODEL_CODE, so different models produce separate index caches and existing indexes remain valid.
New tests cover the override, the explicit model code and the no-env default; the full suite passes (155 passed) and ruff + mypy --strict are clean.
Motivating real-world case: agent memory in a non-English language (Russian), where nomic-embed-text's retrieval quality on paraphrase queries was too weak to be usable.