recall()

recall() takes a natural language query, searches across all stored memories in the profile, and returns a synthesized answer grounded in the actual stored content.

Signature

profile.recall(query: string, options?: RecallOptions): Promise<RecallResult>

Parameters

`query`

A natural language question or search query. Maximum 1 KB (1,024 bytes UTF-8).

Examples:

"What editor does the user prefer?"
"When was the last production deployment?"
"What are the team coding conventions?"
"Summarize recent deployment events"

`options`

type RecallOptions = {
  thinkingLevel?: "low" | "medium" | "high";
  responseLength?: "short" | "medium" | "long";
  referenceDate?: Date | string;
};

Field	Type	Default	Description
`thinkingLevel`	`"low"` \| `"medium"` \| `"high"`	`"low"`	Controls the breadth of retrieval. Higher levels search more candidates but take longer.
`responseLength`	`"short"` \| `"medium"` \| `"long"`	`"medium"`	Controls the verbosity of the synthesized answer.
`referenceDate`	`Date` \| `string`	Current time	A temporal anchor for date-relative queries. Accepts a `Date` object or ISO 8601 string (for example, `"2026-04-21"` or `"2026-04-21T12:00:00Z"`).

Thinking levels

The thinkingLevel option controls how broadly the system searches for relevant memories:

Level	Behavior	Best for
`"low"`	Fastest retrieval. Searches fewer candidates.	Simple factual lookups where speed matters.
`"medium"`	Broader search. Considers more candidates.	Questions that may span multiple memories.
`"high"`	Most thorough. Searches the widest set of candidates.	Complex queries, aggregations, or questions about patterns over time.

Reference date

The referenceDate option anchors temporal reasoning. Use it when the query contains relative time references and "now" should mean something other than the current time.

For example, if you want to ask "What happened last week?" as of a specific date:

const result = await profile.recall("What happened last week?", {
  referenceDate: "2026-04-14",
});

Return value

type RecallResult = {
  answer: string;
};

Field	Type	Description
`answer`	`string`	A natural language answer synthesized from the most relevant memories. Empty string if no relevant memories were found.

Examples

const profile = await env.MEMORY.getProfile("alice");

const result = await profile.recall("What editor does the user prefer?");
console.log(result.answer);
// "The user prefers VS Code as their primary editor."

const profile = await env.MEMORY.getProfile("alice");

const result = await profile.recall("What editor does the user prefer?");
console.log(result.answer);
// "The user prefers VS Code as their primary editor."

With options

JavaScript
TypeScript

const result = await profile.recall(
  "Summarize all deployment events from last month",
  {
    thinkingLevel: "high",
    responseLength: "long",
    referenceDate: "2026-04-01",
  },
);

console.log(result.answer);
// A detailed summary of deployment events from March 2026

const result = await profile.recall("Summarize all deployment events from last month", {
  thinkingLevel: "high",
  responseLength: "long",
  referenceDate: "2026-04-01",
});

console.log(result.answer);
// A detailed summary of deployment events from March 2026

Injecting memory into an LLM call

JavaScript
TypeScript

const profile = await env.MEMORY.getProfile(userId);

// Recall relevant context for the user's question
const memory = await profile.recall(userQuestion, {
  thinkingLevel: "medium",
  responseLength: "short",
});

// Include memory context in the LLM prompt
const response = await env.AI.run("@cf/meta/llama-3.3-70b-instruct-fp8-fast", {
  messages: [
    {
      role: "system",
      content: `You are a helpful assistant. Here is what you know about the user:\n\n${memory.answer}`,
    },
    { role: "user", content: userQuestion },
  ],
});

const profile = await env.MEMORY.getProfile(userId);

// Recall relevant context for the user's question
const memory = await profile.recall(userQuestion, {
  thinkingLevel: "medium",
  responseLength: "short",
});

// Include memory context in the LLM prompt
const response = await env.AI.run("@cf/meta/llama-3.3-70b-instruct-fp8-fast", {
  messages: [
    {
      role: "system",
      content: `You are a helpful assistant. Here is what you know about the user:\n\n${memory.answer}`,
    },
    { role: "user", content: userQuestion },
  ],
});

Limits

Parameter	Limit
Query size	1 KB (1,024 bytes UTF-8)

Refer to Limits for the complete list of constraints.