Use Shared Memory Across AI Agents

Supported agents on the same TermSquad computer can use the shared memory store for the agent Linux user. You can change agents without manually pasting every saved project decision into a new conversation.

Hand off context to another agent

1. Ask the first agent to save important decisions and a concise progress summary.
2. Include the project name, relevant files, completed work, and remaining tasks.
3. Open the second agent in the same repository under the agent user.
4. Ask it to identify the current project and recall its saved context.
5. Have it check the recalled information against the current files before continuing.

For example:

Recall this repository’s latest saved context. Summarize the unfinished task and verify the current Git diff before editing.

How do agents connect?

For compatible integrations, the memory MCP server is named squad_memory. The agent discovers tools for saving, searching, reading context, and retrieving observations.

Pi uses the termsquad memory CLI path. If MCP is unavailable in another tool, the supported CLI commands may provide a fallback.

A tool should say when recall failed instead of presenting guessed history as saved memory.

What is shared?

Selected observations and summaries are shared. Provider conversations, subscriptions, model quotas, and authentication are not merged.

The store is local to this computer. Another computer, a different Linux user, or an unrelated project should not be assumed to have the same context.

See How Shared Memory Works and Shared Memory Isn’t Available.