01 |HYBRID RETRIEVAL
Cheap questions stay cheap. Fast when it can be, thorough when it must be.
Natural language query
Generate embeddings
Provider-pluggable
Entity search
entities
HNSW · COSINE
Vector search
chunks
HNSW · COSINE
Fact chain traversal
entity → relation → entity
3-DEPTH RECURSIVE DFS
memories + facts
02 |WHAT YOU GET
Many signals, one ranker. Meaning, exact terms and structure, weighed together.
Vector recall
Semantic similarity over embeddings stored on the same records.
Read more
Lexical (BM25)
Exact keyword matching, so specific terms survive paraphrase.
Read more
Graph traversal
Connected entities pulled in through typed relationships.
Read more
Personalised PageRank
Importance weighted by how memories connect, as well as how they read.
Read more
Keyword bridges
Shared terms link memories that vector similarity alone would miss.
Read more
Sources on every answer
Recall returns the exact records behind each result, ready to cite.
Read more
03 |WHY IT MATTERS
Why similarity alone falls short. Alike is not the same as true, current or relevant.
01
Tiered, so it stays cheap
Direct lookups answer from the graph; only genuine retrieval runs the full fusion. Cost tracks difficulty rather than volume.
02
Reconciled, not stale
Superseded facts step aside for the current one, so recall reflects what is true now.
03
Provable, every time
Each answer carries the exact records behind it.
GET STARTED
Give your agents recall they can trust. Trustworthy recall, durable memory and a knowledge graph: one system to buy, secure and audit once. Built on SurrealDB.
