Deep Agents Memory

Deep Agents use pluggable backends for file operations and memory:

Short-term (StateBackend): Persists within a single thread, lost when thread ends Long-term (StoreBackend): Persists across threads and sessions Hybrid (CompositeBackend): Route different paths to different backends

FilesystemMiddleware provides tools: ls, read_file, write_file, edit_file, glob, grep

| Use Case | Backend | Why | |----------|---------|-----| | Temporary working files | StateBackend | Default, no setup | | Local development CLI | FilesystemBackend | Direct disk access | | Cross-session memory | StoreBackend | Persists across threads | | Hybrid storage | CompositeBackend | Mix ephemeral + persistent |

Default StateBackend stores files ephemerally within a thread.

from deepagents import create_deep_agent

agent = create_deep_agent()  # Default: StateBackend
result = agent.invoke({
    "messages": [{"role": "user", "content": "Write notes to /draft.txt"}]
}, config={"configurable": {"thread_id": "thread-1"}})
# /draft.txt is lost when thread ends

Default StateBackend stores files ephemerally within a thread.

import { createDeepAgent } from "deepagents";

const agent = await createDeepAgent();  // Default: StateBackend
const result = await agent.invoke({
  messages: [{ role: "user", content: "Write notes to /draft.txt" }]
}, { configurable: { thread_id: "thread-1" } });
// /draft.txt is lost when thread ends

Configure CompositeBackend to route paths to different storage backends.

from deepagents import create_deep_agent
from deepagents.backends import CompositeBackend, StateBackend, StoreBackend
from langgraph.store.memory import InMemoryStore

store = InMemoryStore()

composite_backend = lambda rt: CompositeBackend(
    default=StateBackend(rt),
    routes={"/memories/": StoreBackend(rt)}
)

agent = create_deep_agent(backend=composite_backend, store=store)

# /draft.txt -> ephemeral (StateBackend)
# /memories/user-prefs.txt -> persistent (StoreBackend)

Configure CompositeBackend to route paths to different storage backends.

import { createDeepAgent, CompositeBackend, StateBackend, StoreBackend } from "deepagents";
import { InMemoryStore } from "@langchain/langgraph";

const store = new InMemoryStore();

const agent = await createDeepAgent({
  backend: (config) => new CompositeBackend(
    new StateBackend(config),
    { "/memories/": new StoreBackend(config) }
  ),
  store
});

// /draft.txt -> ephemeral (StateBackend)
// /memories/user-prefs.txt -> persistent (StoreBackend)

Files in /memories/ persist across threads via StoreBackend routing.

# Using CompositeBackend from previous example
config1 = {"configurable": {"thread_id": "thread-1"}}
agent.invoke({"messages": [{"role": "user", "content": "Save to /memories/style.txt"}]}, config=config1)

config2 = {"configurable": {"thread_id": "thread-2"}}
agent.invoke({"messages": [{"role": "user", "content": "Read /memories/style.txt"}]}, config=config2)
# Thread 2 can read file saved by Thread 1

Files in /memories/ persist across threads via StoreBackend routing.

// Using CompositeBackend from previous example
const config1 = { configurable: { thread_id: "thread-1" } };
await agent.invoke({ messages: [{ role: "user", content: "Save to /memories/style.txt" }] }, config1);

const config2 = { configurable: { thread_id: "thread-2" } };
await agent.invoke({ messages: [{ role: "user", content: "Read /memories/style.txt" }] }, config2);
// Thread 2 can read file saved by Thread 1

Use FilesystemBackend for local development with real disk access and human-in-the-loop.

from deepagents import create_deep_agent
from deepagents.backends import FilesystemBackend
from langgraph.checkpoint.memory import MemorySaver

agent = create_deep_agent(
    backend=FilesystemBackend(root_dir=".", virtual_mode=True),  # Restrict access
    interrupt_on={"write_file": True, "edit_file": True},
    checkpointer=MemorySaver()
)

# Agent can read/write actual files on disk

Use FilesystemBackend for local development with real disk access and human-in-the-loop.

import { createDeepAgent, FilesystemBackend } from "deepagents";
import { MemorySaver } from "@langchain/langgraph";

const agent = await createDeepAgent({
  backend: new FilesystemBackend({ rootDir: ".", virtualMode: true }),
  interruptOn: { write_file: true, edit_file: true },
  checkpointer: new MemorySaver()
});

Security: Never use FilesystemBackend in web servers - use StateBackend or sandbox instead.

Access the store directly in custom tools for long-term memory operations.

from langchain.tools import tool, ToolRuntime
from langchain.agents import create_agent
from langgraph.store.memory import InMemoryStore

@tool
def get_user_preference(key: str, runtime: ToolRuntime) -> str:
    """Get a user preference from long-term storage."""
    store = runtime.store
    result = store.get(("user_prefs",), key)
    return str(result.value) if result else "Not found"

@tool
def save_user_preference(key: str, value: str, runtime: ToolRuntime) -> str:
    """Save a user preference to long-term storage."""
    store = runtime.store
    store.put(("user_prefs",), key, {"value": value})
    return f"Saved {key}={value}"

store = InMemoryStore()

agent = create_agent(
    model="gpt-4.1",
    tools=[get_user_preference, save_user_preference],
    store=store
)

What Agents CAN Configure

• Backend type and configuration
• Routing rules for CompositeBackend
• Root directory for FilesystemBackend
• Human-in-the-loop for file operations

What Agents CANNOT Configure

• Tool names (ls, readfile, writefile, edit_file, glob, grep)
• Access files outside virtual_mode restrictions
• Cross-thread file access without proper backend setup

StoreBackend requires a store instance.

# WRONG
agent = create_deep_agent(backend=lambda rt: StoreBackend(rt))

# CORRECT
agent = create_deep_agent(backend=lambda rt: StoreBackend(rt), store=InMemoryStore())

StoreBackend requires a store instance.

// WRONG
const agent = await createDeepAgent({ backend: (c) => new StoreBackend(c) });

// CORRECT
const agent = await createDeepAgent({ backend: (c) => new StoreBackend(c), store: new InMemoryStore() });

StateBackend files are thread-scoped - use same thread_id or StoreBackend for cross-thread access.

# WRONG: thread-2 can't read file from thread-1
agent.invoke({"messages": [...]}, config={"configurable": {"thread_id": "thread-1"}})  # Write
agent.invoke({"messages": [...]}, config={"configurable": {"thread_id": "thread-2"}})  # File not found!

StateBackend files are thread-scoped - use same thread_id or StoreBackend for cross-thread access.

// WRONG: thread-2 can't read file from thread-1
await agent.invoke({ messages: [...] }, { configurable: { thread_id: "thread-1" } });  // Write
await agent.invoke({ messages: [...] }, { configurable: { thread_id: "thread-2" } });  // File not found!

Path must match CompositeBackend route prefix for persistence.

# With routes={"/memories/": StoreBackend(rt)}:
agent.invoke(...)  # /prefs.txt -> ephemeral (no match)
agent.invoke(...)  # /memories/prefs.txt -> persistent (matches route)

Path must match CompositeBackend route prefix for persistence.

// With routes: { "/memories/": StoreBackend }:
await agent.invoke(...);  // /prefs.txt -> ephemeral (no match)
await agent.invoke(...);  // /memories/prefs.txt -> persistent (matches route)

Use PostgresStore for production (InMemoryStore lost on restart).

# WRONG                              # CORRECT
store = InMemoryStore()              store = PostgresStore(connection_string="postgresql://...")

Use PostgresStore for production (InMemoryStore lost on restart).

// WRONG                                    // CORRECT
const store = new InMemoryStore();          const store = new PostgresStore({ connectionString: "..." });

Enable virtual_mode=True to restrict path access (prevents ../ and ~/ escapes).

backend = FilesystemBackend(root_dir="/project", virtual_mode=True)  # Secure

CompositeBackend matches longest prefix first.

routes = {"/mem/": StoreBackend(rt), "/mem/temp/": StateBackend(rt)}
# /mem/file.txt -> StoreBackend, /mem/temp/file.txt -> StateBackend (longer match)

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: Harmeet10000
• Source: Harmeet10000/skills
• License: MIT

Install and usage instructions live in the source repository linked above.