Mehmet EmrahKonyaTR
← Back to glossary

Agent Memory

Ajan Belleği

D3 · Prompts, context, RAG, tools and agents

In this entry, agent memory means application-managed information retained or retrieved from previous interactions so an agent can use it in later work.

Review status: 2026-12-04

Technical explanation

LangChain and Google ADK distinguish information for one conversation from information retrieved across conversations; their storage interfaces differ.

Conceptual boundaries

This entry concerns how an application retains and reuses information, not computer RAM capacity. Retained history need not all be supplied to the current model call: the context window limits how much can be supplied at once.

Provider-neutral example

An assistant remembers a user's preferred date format for future reports.

Limitations

Remembered material can be stale, incomplete, wrongly attributed, unavailable, or inappropriate to reuse. Persistence does not make it true, authorized, or suitable for the next request.

Related concepts

Atomic claims and evidence

  1. 1.1LangChain Docs: Short-term memory
    Source
    LangChain Docs: Short-term memory
    Source role
    Authoritative source
    Exact locator
    Overview
    Supported claim
    LangChain defines agent memory as a system that remembers information about previous interactions and says short-term memory can retain those interactions within one thread or conversation.
    Last verification
    Review due
    Scope limitation
    This is LangChain documentation terminology; it is not a universal storage architecture or a guarantee of recall accuracy.
  2. 2.1LangChain Docs: Short-term memory
    Source
    LangChain Docs: Short-term memory
    Source role
    Authoritative source
    Exact locator
    Overview: short-term and long-term memory
    Supported claim
    LangChain distinguishes thread-level short-term memory from long-term memory that stores and recalls user-specific or application-level data across threads and sessions.
    Last verification
    Review due
    Scope limitation
    The distinction describes a common design scope; it does not prescribe a database, retention period, or privacy policy.
  3. 3.1LangChain Docs: Short-term memory
    Source
    LangChain Docs: Short-term memory
    Source role
    Authoritative source
    Exact locator
    Overview: context-window limits and removing stale information
    Supported claim
    LangChain says a full conversation history may not fit an LLM context window and that applications can remove or forget stale information because context windows are limited.
    Last verification
    Review due
    Scope limitation
    This is LangChain’s documented design consideration, not a claim that forgetting is always correct or that an application’s retained memory is accurate.
  4. 4.1Google ADK: Memory — Long-term knowledge with MemoryService
    Source
    Google ADK: Memory — Long-term knowledge with MemoryService
    Source role
    Authoritative source
    Exact locator
    Opening distinction between Session/State and long-term knowledge; The MemoryService role
    Supported claim
    Google ADK separates session history and state from a MemoryService for searchable information across interactions.
    Last verification
    Review due
    Scope limitation
    This independently supports the scope distinction, not identical storage behavior or universal terminology.