RECALL

Recall you can trust

Every turn, agent memory feeds the context window the slice that matters - and recall is engineered for correctness. A hybrid, multi-signal ranker weighs meaning, exact keywords, and graph connections together, and every answer comes back with the exact sources behind it.

Join the waitlist Technical deep-dive

01 |HYBRID RETRIEVAL

Cheap questions stay cheap

A typed lookup answers straight from the graph and never touches the ranker. Only real retrieval runs the full fusion, with a broader sweep held in reserve for when the answer is thin - so recall is fast when it can be and thorough when it must be.

Natural language query

Generate embeddingsProvider-pluggable

Vector searchchunks

HNSW · COSINE

Entity searchentities

HNSW · COSINE

Fact chain traversalentity → relation → entity

3-depth recursive DFS

Memories + facts

// npm install @surrealdb/agent-memory
import { AgentMemory } from "@surrealdb/agent-memory";

const memory = new AgentMemory({ context: "acme-prod" });

await memory.remember("I work at Acme as CTO");
const hits = await memory.recall("what do I do at Acme");

Many signals, fused in one ranker

Meaning, exact terms, and graph structure are weighed together in a single ranker - not vector similarity alone, and not a rerank bolted on afterwards.

Vector recall

Semantic similarity over embeddings stored on the same records.

Learn more

Lexical (BM25)

Exact keyword matching, so specific terms are never lost to paraphrase.

Learn more

Graph traversal

Connected entities pulled in through typed relationships.

Learn more

Personalised PageRank

Importance weighted by how memories connect, not just how they read.

Learn more

Keyword bridges

Shared terms link memories that vector similarity alone would miss.

Learn more

Sources on every answer

Recall returns the exact records behind each result, ready to cite.

Learn more

02 |ONE SUBSTRATE

Similarity alone gets the wrong thing

A pure vector search returns what reads similarly - which is not always what's true, current, or relevant. Agent Memory fuses multiple signals and reconciles what changed, so recall surfaces the right thing, with its sources.