AgentStore Notes Agent

This folder contains an agent example that behaves like the default tool calling agent, but uses the AgentStore to remember key details that persist between conversations.

At a high level, this template shows how to:

Read a value from ram_client.store before running a query.
Write an updated value to ram_client.store after running a query.

Input to This Agent Experiment

Retrieval Settings - System Prompt

Any system prompt works for this template.

Tools

Any MCP tools configured in RAM will work with this template.

Collections

Any RAM collection will work with this template.

Environment Variables (optional)

No environment variables are required for this example.

Runtime Behavior

For each user message:

The template fetches any notes saved in the AgentStore from previous turns.
If notes exist, they’re added alongside the user’s message before it’s sent to the agent.
The template invokes the agent and gets the final response.
The template asks the LLM to summarize the exchange into an updated set of notes (a direct LLM query, without tools or retrieval), and saves those notes back to the AgentStore with ram_client.store.set(...).

Debug Commands

These commands read/write the store directly, bypassing the LLM, so you can verify what is persisted:

#shownotes: Return the raw notes currently stored in the AgentStore.
#clearnotes: Clear the stored notes.
#compactnotes: Condense the stored notes if they’ve grown too large, keeping the key facts, preferences, and decisions but trimming everything else.

Matching ignores case, spaces, and a leading #, so show notes, ShowNotes, and #shownotes all work the same way.

Note on Persistent Storage

Within a single conversation, RAM already replays the full session history to the LLM on every turn (ram_client.get_session_history_as_langchain()). This happens for most default templates, including default_tool_calling_agent.

What AgentStore adds is memory that persists outside the current conversation:

It persists across brand-new conversations/sessions with the same agent.
It stays a small, bounded summary instead of growing with every turn of raw transcript.

Automation Hook (Optional)

There are no automations for this agent.