Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.hud.ai/llms.txt

Use this file to discover all available pages before exploring further.

Memory tools let agents store and retrieve information that persists beyond a single request. Options include file-based storage (Claude’s native memory), session-based key-value storage, and semantic search.

​
Quick Reference

ToolAgentStoragePersistence
ClaudeMemoryToolClaudeFiles in /memoriesAcross conversations
SessionMemoryToolAnyIn-memory dictSession only
GeminiMemoryToolGeminiIn-memory dictSession only
All memory tools are in the hud.tools.memory module: 
from hud.tools.memory import (
    ClaudeMemoryTool,
    SessionMemoryTool,
    GeminiMemoryTool,
)

​
ClaudeMemoryTool

File-based memory for Claude. Uses native memory_20250818 API. Stores files in a /memories directory. 
from hud.tools.memory import ClaudeMemoryTool

memory = ClaudeMemoryTool(memories_dir="/memories")
Commands: view, create, str_replace, insert, delete, rename 
# View memories directory
await memory(command="view", path="/memories")

# Create a memory file
await memory(
    command="create",
    path="/memories/user_prefs.md",
    file_text="# Preferences\n\n- Theme: dark\n- Language: Python",
)

# Update memory
await memory(
    command="str_replace",
    path="/memories/user_prefs.md",
    old_str="- Theme: dark",
    new_str="- Theme: light",
)

# View file contents
await memory(command="view", path="/memories/user_prefs.md")

# Delete
await memory(command="delete", path="/memories/old_notes.md")

# Rename/move
await memory(
    command="rename",
    old_path="/memories/temp.md",
    new_path="/memories/archive/temp.md",
)
Paths must start with /memories. Directory traversal is blocked.

​
SessionMemoryTool

Simple key-value memory for any agent. Stores data in an in-memory dictionary. 
from hud.tools.memory import SessionMemoryTool

memory = SessionMemoryTool()
Actions: add, query, list 
# Store memory with a key
await memory(action="add", key="user_lang", value="Python is their preferred language")

# Query by key
result = await memory(action="query", key="user_lang")

# List all keys
result = await memory(action="list")
Useful for simple session context that doesn’t need semantic search or persistence.

​
GeminiMemoryTool

Gemini CLI-style memory with read/write operations. Uses in-memory storage. 
from hud.tools.memory import GeminiMemoryTool

memory = GeminiMemoryTool()
Actions: read, write, list 
# Write memory
await memory(action="write", key="context", value="User is working on a web app")

# Read memory
result = await memory(action="read", key="context")

# List all memories
result = await memory(action="list")

​
When to Use Which

Use CaseTool
Claude with native APIClaudeMemoryTool
Structured file storageClaudeMemoryTool
Simple key-value storageSessionMemoryTool
Gemini agentsGeminiMemoryTool

​
Typical Setup

For Claude: 
from hud import Environment
from hud.tools import BashTool, EditTool
from hud.tools.memory import ClaudeMemoryTool

env = Environment("claude-env")
env.add_tool(BashTool())
env.add_tool(EditTool())
env.add_tool(ClaudeMemoryTool())
For Gemini: 
from hud import Environment
from hud.tools.coding import GeminiShellTool, GeminiEditTool
from hud.tools.memory import GeminiMemoryTool

env = Environment("gemini-env")
env.add_tool(GeminiShellTool())
env.add_tool(GeminiEditTool())
env.add_tool(GeminiMemoryTool())
For any agent with simple memory: 
from hud import Environment
from hud.tools import BashTool
from hud.tools.memory import SessionMemoryTool

env = Environment("generic-env")
env.add_tool(BashTool())
env.add_tool(SessionMemoryTool())

​
Custom Memory

Key-value storage: 
from hud.tools import BaseTool
from mcp.types import ContentBlock, TextContent

class ContextTool(BaseTool):
    def __init__(self):
        super().__init__(name="context", description="Store and retrieve context")
        self._store: dict[str, str] = {}
    
    async def __call__(
        self, action: str, key: str, value: str | None = None
    ) -> list[ContentBlock]:
        if action == "set" and value:
            self._store[key] = value
            return [TextContent(text=f"Stored: {key}", type="text")]
        elif action == "get":
            val = self._store.get(key, "Not found")
            return [TextContent(text=val, type="text")]
        elif action == "list":
            keys = ", ".join(self._store.keys()) or "Empty"
            return [TextContent(text=keys, type="text")]
        return [TextContent(text="Unknown action", type="text")]