Memory Tools

5 tools that let LLM agents manage their own memory via tool calling.

Setup


from unforget import MemoryStore, MemoryToolExecutor
 
store = MemoryStore("postgresql://...")
await store.initialize()
 
executor = MemoryToolExecutor(store, org_id="acme", agent_id="bot")
 
# Get tool schemas for your LLM
openai_tools = executor.to_openai()
anthropic_tools = executor.to_anthropic()
generic_tools = executor.to_generic()

Tools

memory_store

Save a fact to memory.


{
  "content": "User prefers dark mode",
  "memory_type": "insight",
  "tags": ["preference", "ui"],
  "importance": 0.7
}

memory_search

Recall relevant memories.


{
  "query": "user preferences",
  "limit": 5
}

memory_list

Browse memories by type or tags.


{
  "memory_type": "insight",
  "tags": ["deploy"],
  "limit": 20
}

memory_forget

Remove a memory.


{
  "memory_id": "uuid-string"
}

memory_update

Update or supersede a memory.


{
  "memory_id": "uuid-string",
  "new_content": "User now prefers light mode"
}

Executing tool calls


# Check if a tool call is a memory tool
if executor.is_memory_tool_call(tool_name):
    result = await executor.execute(tool_name, tool_args)
 
# Or handle full LLM responses
tool_messages = await executor.handle_openai_response(response)
tool_messages = await executor.handle_anthropic_response(response)