Skip to main content

Agent Memory

Agent Memory provides MCP-based persistent file memory for AI Agents. It enables Agents to retain context across sessions by storing structured memories as files, accessible via the Model Context Protocol (MCP).

Overview​

AI Agents typically lose all context between sessions. Agent Memory solves this by providing:

  • Persistent Storage — memories survive across Agent restarts and sessions
  • File-based Architecture — memories stored as structured files for transparency and portability
  • MCP Interface — standard Model Context Protocol server with 30+ tools for seamless Agent integration
  • Sandboxed Execution — operates safely within restricted environments

Prerequisites​

  • Linux (x86_64 or aarch64)
  • An MCP-compatible Agent runtime

Installation​

anolisa install agent-memory

Option 2: YUM (Alinux, requires ANOLISA YUM repo)​

sudo yum install agent-memory

Option 3: Source build (developers)​

cd src/agent-memory && make build

Quick Start​

# 1. Install Agent Memory
anolisa install agent-memory

# 2. Start the MCP server
agent-memory serve

# 3. Configure your Agent runtime to connect to the MCP server
# (see Integration section below)

Integration​

Agent Memory runs as an MCP server. Configure your Agent runtime to connect:

{
  "mcpServers": {
    "agent-memory": {
      "command": "agent-memory",
      "args": ["serve"]
    }
  }
}

The Agent can then use MCP tools to read/write memories during conversation.

MCP Tools​

Agent Memory exposes 30+ MCP tools. Key categories:

File Operations​

  • mem_read / mem_write / mem_append / mem_edit — read, write, append, and edit memory files
  • mem_list / mem_grep / mem_diff — list, search, and diff memory content
  • mem_mkdir / mem_remove — manage memory directories and files
  • mem_promote — promote a memory entry

Session & Context​

  • mem_session_log — log session activity
  • memory_search / memory_observe / memory_get_context — semantic search and context retrieval
  • memory_sessions / memory_timeline / memory_summary — session history and summaries

Maintenance​

  • mem_dream / mem_consolidate / mem_compact — background consolidation and compaction
  • mem_index_refresh — refresh the memory index
  • mem_snapshot / mem_snapshot_list / mem_snapshot_restore — snapshot management
  • mem_log / mem_revert — history log and revert

Task Management​

  • memory_task_save / memory_task_resume / memory_task_list / memory_task_close — save and resume multi-step tasks

Import/Export & Meta​

  • mem_export / mem_import — bulk export and import
  • memory_about / memory_forget / memory_auto_created / memory_consent — metadata and controls

Configuration​

Configuration file: ~/.anolisa/memory.toml

This file is optional and is not auto-generated. When absent, Agent Memory uses built-in defaults. Create it only if you need to override default behavior.

# Example: override defaults
[storage]
path = "~/.anolisa/memory/"

[server]
transport = "stdio"

Data Directory​

Memory files are stored in ~/.anolisa/memory/ by default.

FAQ​

Q: Where are memories stored? A: By default in ~/.anolisa/memory/ as structured files.

Q: Is a config file required? A: No. Agent Memory works with built-in defaults. The optional config at ~/.anolisa/memory.toml is only needed to override specific settings.

Q: Can Agent Memory work in sandboxed environments? A: Yes. Agent Memory is designed to operate within restricted/sandboxed execution contexts.

Q: How does this differ from Tokenless? A: Tokenless compresses in-context information to save Tokens. Agent Memory offloads knowledge to persistent storage so it doesn't need to be in-context at all.