v0.7.1

CtxVault v0.7.1

CtxVault v0.7.1 lets you choose which sentence-transformers embedding model each semantic vault uses, configured once at init time and pinned for all indexing and querying on that vault.

---

Added

Per-vault embedding model selection — ctxvault init now accepts --embedding-model <model> for semantic vaults. The chosen model is stored in the vault's config.json and used consistently by both index and query, ensuring vectors always stay comparable. A process touching multiple vaults loads each model exactly once via a per-name cache. Omitting the flag preserves the existing default (all-MiniLM-L6-v2), so all existing vaults continue to work without any migration.

ctxvault init my-vault --embedding-model all-mpnet-base-v2

Passing --embedding-model on a skill vault is rejected immediately with a clear error, as skill vaults do not use embeddings.

---

Tests

Full coverage for #18 — new tests in test_core.py and test_cli.py assert that the model is stored in config when overridden and absent when using the default, that --embedding-model on a skill vault is rejected, that both index and query thread the vault's configured model to embed_list, and that get_model caches one instance per model name and falls back to the default correctly.

---

What's unchanged

All existing CLI commands, API endpoints, MCP tools, and vault operations work exactly as before.