Memories

​

What is a Memory?

• Content: the extracted information
• Metadata: additional context and tags
• Entities: extracted people, places, technologies
• Embeddings: semantic representations for search

​

Creating memories

​

From text

client.memories.add(
    space_id="space_123",
    content="User prefers dark mode and concise responses"
)

await client.memories.add({
  space_id: 'space_123',
  content: 'User prefers dark mode and concise responses',
});

​

From files

with open("meeting_notes.pdf", "rb") as file:
    client.upload.batch.create(
        space_id="space_123",
        files=[file]
    )

const file = fs.createReadStream('meeting_notes.pdf');
await client.upload.batch.create({
  space_id: 'space_123',
  files: [file],
});

​

From videos

with open("team_meeting.mp4", "rb") as video:
    client.upload.batch.create(
        space_id="space_123",
        files=[video]
    )

const video = fs.createReadStream('team_meeting.mp4');
await client.upload.batch.create({
  space_id: 'space_123',
  files: [video],
});

• Spoken dialogue (transcription)
• Visual context (on-screen content)
• Text (slides, captions)
• Temporal information

​

Memory lifecycle

1. Creation: Add via API or upload files
2. Indexing: Entities extracted, graph built, embeddings created
3. Retrieval: Search by semantic meaning and graph connections
4. Updates: New content extends the knowledge graph
5. Deletion: Remove when no longer needed

​

Use cases

​

User preferences

client.memories.add(
    space_id=f"user_{user_id}",
    content="User prefers technical explanations",
    metadata={"type": "preference"}
)

​

Conversation history

client.memories.add(
    space_id="conversation_123",
    content=f"User: {user_message}\nAssistant: {assistant_response}"
)

​

Knowledge bases

# Upload documentation
with open("api_docs.pdf", "rb") as file:
    client.upload.batch.create(space_id="docs", files=[file])

# Search later
results = client.memories.search(
    space_id="docs",
    query="How do I authenticate API requests?"
)

​

Best practices

• Add rich metadata for better filtering
• Use clear, self-contained content
• Organize related memories in the same space
• Include entity relationships in content
• Update when information changes

​

Next steps