SDK: Memory

Overview

The Memory API manages user/agent/app memories and vector store search. It provides sync and async clients with consistent method shapes.

Exports:

• FlotorchMemory, FlotorchAsyncMemory – CRUD + search for memories
• FlotorchVectorStore, FlotorchAsyncVectorStore – semantic search in a vector store

---

Setup

from flotorch.sdk.memory import (
  FlotorchMemory,
  FlotorchAsyncMemory,
  FlotorchVectorStore,
  FlotorchAsyncVectorStore,
)

API_KEY = "<your_api_key>"
BASE_URL = "https://gateway.flotorch.cloud"
PROVIDER = "<flotorch_memory_provider>"  # FloTorch Memory Provider configured in the FloTorch Console
VECTORSTORE_ID = "<your_vectorstore_id>"

---

Memory clients

FlotorchMemory

FlotorchMemory(
  api_key: str,
  base_url: str,
  provider_name: str,
)

add(…) -> dict

add(
  messages: list[dict],
  userId: str | None = None,
  agentId: str | None = None,
  appId: str | None = None,
  sessionId: str | None = None,
  metadata: dict | None = None,
  timestamp: str | None = None,
  providerParams: dict | None = None,
) -> dict

• Adds one or more chat-style messages to memory.

get(…) -> dict

get(memory_id: str) -> dict

• Fetches a memory by id.

update(…) -> dict

update(
  memory_id: str,
  content: str | None = None,
  metadata: dict | None = None,
) -> dict

• Updates memory content and/or metadata.

delete(…) -> dict

delete(memory_id: str) -> dict

• Deletes a memory by id.

search(…) -> dict

search(
  userId: str | None = None,
  agentId: str | None = None,
  appId: str | None = None,
  sessionId: str | None = None,
  createFrom: str | None = None,
  createTo: str | None = None,
  updateFrom: str | None = None,
  updateTo: str | None = None,
  categories: list[str] | None = None,
  metadata: dict | None = None,
  page: int | None = 1,
  limit: int | None = 20,
  query: str | None = None,
) -> dict

• Searches memories using rich filters. Returns provider-shaped payload.

---

FlotorchAsyncMemory

Constructor:

FlotorchAsyncMemory(
  api_key: str,
  base_url: str,
  provider_name: str,
)

Async equivalents of add, get, update, delete, and search are provided with the same parameters and return shapes.

---

Vector store clients

FlotorchVectorStore

Constructor:

FlotorchVectorStore(
  base_url: str,
  api_key: str,
  vectorstore_id: str,
)

search(…) -> dict

search(
  query: str,
  max_number_of_result: int | None = None,
  ranker: str | None = None,
  score_threshold: float | None = None,
  rewrite_query: bool | None = None,
) -> dict

• Runs semantic search. Optional parameters act as overrides.

---

FlotorchAsyncVectorStore

Constructor:

FlotorchAsyncVectorStore(
  base_url: str,
  api_key: str,
  vectorstore_id: str,
)

Async search with the same parameters as the sync client.

---

Notes

• Provide valid provider_name configured in your FloTorch deployment.