ONNX embedding models run locally via Xenova
ONNX embedding models run locally via Xenova Transformers
Recall
Semantic Search
Local embeddings — via Xenova Transformers (ONNX), llama.cpp (GGUF), or a remote embedding API — so AI clients can recall memories by meaning, not just keywords.
Overview
What Semantic Search does
Local embeddings — via Xenova Transformers (ONNX), llama.cpp (GGUF), or a remote embedding API — so AI clients can recall memories by meaning, not just keywords.
Keyword search misses the memory you wrote three months ago using different words. Semantic recall finds it by meaning, which is how your AI tools think about retrieval anyway.
SearchSimilarity
Why it matters
The payoff for your AI memory
GGUF embedding models run via llama.cpp for
GGUF embedding models run via llama.cpp for higher quality on capable machines
Remote embedding API option if you want
Remote embedding API option if you want to offload
Semantic recall exposed to AI clients through
Semantic recall exposed to AI clients through the MCP `vault_search` tool
How it works
From first launch to reusable memory
- 1
Open search in the app or have a connected AI client query the vault.
- 2
Enter a natural-language description of the memory you want, not just keywords.
- 3
Review results ranked by conceptual similarity and open the closest match.
- 4
Rephrase to steer toward a different angle of the same topic when needed.
FAQ
Common questions
Does semantic search require GPU?
No. The default ONNX model runs on CPU. GGUF models can use Metal on macOS or CUDA on Linux if you've set up llama.cpp accordingly.
What happens when I change embedding models?
1AIVault triggers a background backfill that re-embeds every entry with the new model. The vault stays usable while it runs; search quality blends old and new until the backfill completes.
Related
More Recall features
Dashboard & Activity
Dashboard view shows a vault summary plus a live activity feed of every read, write, and classification across all connected AI tools. Deep links via `aivault://` open the app from anywhere.
Learn moreLast Used Stripe on Every Memory
Every entry detail header shows 'Last used X ago by Source · Nx this week' so you know which AI tools are touching a memory and how often.
Learn moreFull-text search with ranking
Text search uses ranked matching so exact terms, titles, sources, and memory bodies are still easy to audit.
Learn moreLocal memory, shared everywhere
Give every AI tool the same memory.
Start free, import real conversations, and reuse your memory across every AI agent you already use.