Memory class

Memory.add, Memory.search, Memory.update, and Memory.delete.

Constructor

const memory = new Memory(agentId: string, config: MemoryConfig);

agentId becomes sourceAgent on all claims written by this instance.

Methods

`add(messages, conversationId)`

Extract → classify → store. Returns { added: Claim[] } (empty if all facts were NOOP).

const { added } = await memory.add(
  [{ role: "user", content: "I'm in PST" }],
  "conv-1",
);

`search(query, opts?)`

Semantic search scoped to this agent's sourceAgent.

interface SearchOpts {
  limit?: number;
  entity?: string;
  conversationId?: string;
  status?: "active" | "superseded" | "disputed";
}

`update(claimId, text)`

Re-extract embedding and update claim text for an owned claim.

`delete(claimId)`

Remove claim row and vector index entry.

Cross-agent search

memory.search() filters by agent. For all agents:

import { search, resolveEmbed, resolveVectorStore } from "vault-memory";

const vectorStore = resolveVectorStore({ db });
const embed = resolveEmbed({ apiKey: process.env.OPENAI_API_KEY });

const results = await search(db, vectorStore, embed, "timezone", {
  status: "active",
  limit: 10,
});

Lower-level exports

Export	Use
`extractFacts`	Custom pipelines without full `add`
`classifyAgainstOwnHistory`	Custom classify flow
`classifyAgainstOtherAgents`	Custom cross-agent flow
`applyOp` / `applyCrossAgentResolution`	Direct op application