autourgos-buffer-memory

In-memory short-term buffer for Autourgos agents. Three classes — a message-count bounded ring buffer, an unbounded conversation buffer, and a TTL-expiring buffer. Fast, zero I/O, ideal for single-session use.

from autourgos_buffer_memory import RuntimeShortTermMemory
from autourgos_agent import Agent
from autourgos_openaichat import OpenAIChatModel

my_llm = OpenAIChatModel(model="gpt-4o-mini")
memory = RuntimeShortTermMemory(max_messages=20)
agent  = Agent(llm=my_llm, memory=memory)

agent.invoke("My name is Jitin")
agent.invoke("What is my name?")
# → "Your name is Jitin."

Features

• RuntimeShortTermMemory — keeps the last N messages in RAM, oldest dropped when the cap is exceeded
• ConversationBufferMemory — same shape, no truncation, keeps every message for the session
• ExpiringBufferMemory — messages carry a time-to-live and are purged once expired; for a long-running/background agent's temporary, run-scoped facts that shouldn't outlive the run
• Implements autourgos_memory.BaseMemory — drop-in for Agent(memory=...)
• Zero I/O, fastest option in the memory family

Table of Contents

• Install
• Classes
• Parameters
• API
• License

Install

pip install autourgos-buffer-memory

Classes

RuntimeShortTermMemory

Keeps the last N messages in RAM. Oldest messages are dropped when the cap is exceeded.

from autourgos_buffer_memory import RuntimeShortTermMemory
from autourgos_agent import Agent
from autourgos_openaichat import OpenAIChatModel

my_llm = OpenAIChatModel(model="gpt-4o-mini")
memory = RuntimeShortTermMemory(max_messages=20)
agent  = Agent(llm=my_llm, memory=memory)

agent.invoke("My name is Jitin")
agent.invoke("What is my name?")
# → "Your name is Jitin."

ConversationBufferMemory

Same as RuntimeShortTermMemory but with no truncation — keeps every message for the session.

from autourgos_buffer_memory import ConversationBufferMemory

memory = ConversationBufferMemory()
agent  = Agent(llm=my_llm, memory=memory)

For long conversations, use autourgos-summary-memory or autourgos-token-memory to stay within context window limits.

ExpiringBufferMemory

Each message carries a time-to-live; expired messages are purged automatically (lazily, on the next add/read — no background thread) and never appear in get_messages()/format_for_llm(). Meant for a long-running or background agent's temporary, run-scoped facts — worth remembering for the next few minutes or hours of a task, but that shouldn't silently persist the way an unbounded buffer would.

from autourgos_buffer_memory import ExpiringBufferMemory

memory = ExpiringBufferMemory(default_ttl_seconds=3600)  # 1 hour default
agent  = Agent(llm=my_llm, memory=memory)

memory.add_user_message("Skip the venv folder for this run.")          # expires in 1 hour
memory.add_user_message("The deploy target is us-east-1.", ttl_seconds=None)  # never expires

Pass ttl_seconds= to any add_*_message() call to override default_ttl_seconds for that one message; ttl_seconds=None (explicit) makes that message permanent even with a default TTL set.

Parameters

RuntimeShortTermMemory

ConversationBufferMemory

ExpiringBufferMemory

API

memory.add_user_message("Hello")
memory.add_agent_message("Hi there!")
memory.add_tool_message("search", "Found 5 results")
memory.add_system_message("You are a helpful assistant")

messages = memory.get_messages()   # list of role/content dicts
context  = memory.format_for_llm() # formatted string for LLM prompt
memory.clear()

License

Apache License 2.0, Copyright (c) 2026 Jitin Kumar Sengar