Memory

​

What is memory?

• Persistent context - Information saved across sessions
• Knowledge building - Accumulate insights over time
• Personalization - Scira remembers your preferences and history
• Retrieval - Search through saved memories semantically

​

Memory vs. chat history

​

How memory works

1. Save - You or Scira saves important information
2. Embedding - Content is converted to semantic vectors
3. Indexing - Stored with your unique user tags
4. Retrieval - Searched when relevant to your queries

// Memory tools powered by Supermemory
import { supermemoryTools } from '@supermemory/tools/ai-sdk';

export function createMemoryTools(userId: string) {
  return supermemoryTools(serverEnv.SUPERMEMORY_API_KEY, {
    containerTags: [userId],
  });
}

​

Using memory

​

Saving memories

"Remember that I prefer TypeScript over JavaScript"
"Save that my favorite color is blue"
"Keep in mind that I work in Pacific timezone"

​

Retrieving memories

"What do you remember about my color preferences?"
"Recall what I told you about Project Apollo"
"What did I save about JWT tokens?"

​

Memory tools

​

Add memory tool

export type AddMemoryTool = Tool<
  {
    memory: string; // The information to save
  },
  | {
      success: boolean;
      memory: any; // Saved memory object
      error?: undefined;
    }
  | {
      success: boolean;
      error: string;
      memory?: undefined;
    }
>;

const result = await addMemory({
  memory: "User prefers dark mode in all applications"
});

​

Search memory tool

export type SearchMemoryTool = Tool<
  {
    informationToGet: string; // What to search for
  },
  | {
      success: boolean;
      results: any[];
      count: number;
      error?: undefined;
    }
  | {
      success: boolean;
      error: string;
      results?: undefined;
      count?: undefined;
    }
>;

const result = await searchMemory({
  informationToGet: "user color preferences"
});
// Returns: { success: true, results: [...], count: 3 }

​

Use cases

​

Personal assistant

• Daily schedule and habits
• Contact information
• Favorite tools and resources
• Personal goals and tracking

​

Research companion

• Key findings and insights
• Reference materials
• Paper summaries
• Research questions and hypotheses

​

Project context

• Team members and roles
• API endpoints and credentials
• Architecture decisions
• Deployment procedures

​

Learning and development

• Programming patterns
• Framework documentation snippets
• Code examples
• Best practices

​

Implementation details

​

User isolation

const containerTags = [userId];

​

Semantic search

• Not keyword matching - Finds conceptually similar content
• Context-aware - Understands relationships between ideas
• Ranking - Results sorted by semantic similarity

​

Memory structure

interface Memory {
  content: string;           // The saved information
  userId: string;            // Owner identifier
  createdAt: Date;           // When it was saved
  metadata?: {               // Optional context
    source?: string;         // Where it came from
    tags?: string[];         // User-defined tags
    [key: string]: any;
  };
}

​

Privacy and security

​

Data protection

• Encrypted storage - All memories encrypted at rest
• User isolation - Container tags prevent cross-user access
• API security - Supermemory API key required for all operations
• No sharing - Memories are never shared with other users or services

​

What gets stored

• Text content - The information you save
• Timestamps - When memories were created
• User tags - Your unique identifier
• Embeddings - Semantic vectors for search (not human-readable)

​

What doesn’t get stored

• Sensitive credentials - Don’t save passwords or API keys
• PII without consent - Be mindful of personal information
• Temporary data - Use chat history for session-specific context

​

Supermemory integration

import { supermemoryTools } from '@supermemory/tools/ai-sdk';

​

Configuration

# .env.local
SUPERMEMORY_API_KEY=your_api_key_here

​

Tool initialization

const memoryTools = createMemoryTools(userId);

// Returns:
// - addMemory tool
// - searchMemory tool

​

AI SDK integration

import { generateText } from 'ai';

const { text } = await generateText({
  model: yourModel,
  tools: {
    ...createMemoryTools(userId),
  },
  prompt: "Remember that I prefer dark mode",
});

​

Best practices

​

Effective memory saving

1. Be specific - “Prefers React over Vue” vs. “Likes React”
2. Include context - “For Project Apollo: deadline is March 15”
3. Use natural language - Save in complete sentences
4. Avoid duplication - Update existing memories instead of creating duplicates

​

Efficient retrieval

1. Ask semantically - “What’s my coding style?” retrieves language preferences
2. Use context clues - Reference project names, topics, or timeframes
3. Be patient - Initial retrieval may take 1-2 seconds
4. Verify results - Check that retrieved info is still current

​

Memory organization

1. Categorize mentally - Think of memories in topics or domains
2. Update regularly - Keep information current
3. Archive old info - Clear outdated memories
4. Use descriptive saves - Make memories searchable

​

Troubleshooting

​

Memories not saving

• Check API key - Ensure SUPERMEMORY_API_KEY is set
• Verify user ID - Must be logged in with valid session
• Network issues - Supermemory API must be reachable

​

Poor retrieval results

• Too vague - Make queries more specific
• No relevant memories - Save information first
• Semantic mismatch - Try rephrasing the query

​

Duplicate memories

• Manual deduplication - Search and remove duplicates
• Better save habits - Check before saving new info
• Update instead - Modify existing memories when appropriate

​

Limitations

• Text only - Cannot store images, files, or binary data
• Search latency - Semantic search takes 1-2 seconds
• No deletion API - Currently cannot delete individual memories
• Context window - Very large memory sets may impact performance

​

API reference

• createMemoryTools(userId) - Initialize memory tools for user
• SearchMemoryTool - Tool interface for searching memories
• AddMemoryTool - Tool interface for adding memories

​

Future enhancements

• Memory management UI - View, edit, and delete memories
• Automatic categorization - AI-organized memory tags
• Memory expiry - Time-based memory lifecycle
• Export/import - Backup and restore memories
• Shared memories - Team knowledge bases (Pro Teams)