Agent Memory Stack

Your agent. Your context.

Agent Memory Stack

Your AI subscription, with a memory stack you run yourself.

Bring persistent memory, reusable skills, Wiki and CodeGraph to your coding agent. Connect your existing model accounts and access project knowledge through remote MCP, from your laptop or your own server.

Interactive Agent Memory Stack setup in a terminal.
Configure the complete stack interactively, save the native settings, and apply them when the container engine is ready.
Agent Memory Stack overview: MemoryProxy and MCP are separate agent connections; Core and Knowledge choose shared CLIProxyAPI or an external model API.
Two agent connections: MemoryProxy for model requests with memory, and MCP for Knowledge tools. Open the image to explore the interactive stack map.

Explore the architecture

The big picture

Stack overview

Explore the complete stack, unchanged TDAI services, and the choice of CLIProxyAPI or an external API for internal models.

Explore the stack A closer look

Inside the MCP bridge

Follow authenticated HTTP through the official MCP SDK to TDAI's stock stdio tools, with a separate process for each HTTP POST.

Explore MCP

Set up your stack

  1. Run ams and select Configure stack. Choose the model source and the Core and Knowledge models.
  2. Keep upstream templates in ~/.config/agent-memory-stack/defaults/ and your settings in overrides/. Configure fills the settings it knows; you control the remaining native options.
  3. Run ams apply to apply the saved files and start the complete stack. Use Show connection details for addresses and credentials.

TDAI updates refresh the upstream defaults while preserving your overrides. AMS supplies configuration and transport integration; TDAI runs with its native behavior.