Julep

Memory Store from Julep is a shared memory layer for individuals and teams using AI agents. It syncs conversations, notes, and decisions across tools like Claude, Codex, Slack, and related MCP clients.

AI Team Collaboration AI Notes Assistant AI Knowledge Management

Visit Website

Overview

Memory Store is a shared memory product for AI tools and teammates. It turns conversations, snippets, notes, and decisions into connected memory so context is not trapped inside individual chats or apps.

The site positions the product as part of your existing workflow rather than a separate destination. It runs as an MCP inside the agents you already use, with examples across Claude, Codex, ChatGPT, Cursor, Slack, Raycast, Linear, Granola, Fathom, and OpenCode.

Features

Syncs from existing work tools

Turns work already happening in tools like Slack, Fathom, and Granola into connected memory that captures what happened, who was involved, and what changed.

Background organization

Groups conversations by project, person, and decision in the background so teams do not need to manually file notes or threads.

MCP-based recall across agents

Runs as an MCP inside the agents you already use, letting you record information in one agent and recall it in another.

Reusable context for playbooks

Supports living playbooks and other reusable context that teammates can read later instead of searching through old threads.

Personal and team memory

Positions memory for both a single user and a whole team, with pages built from your own work and updated as you work.

Fast onboarding path

Uses a Book Demo flow for teams and says setup can be completed in under five minutes.

Use cases

Personal context wiki
An individual using Claude can keep notes, preferences, and context in one readable place that updates as they work.
Decision tracking for teams
A team can capture decisions from meetings and discussions, then refer back to the thread that explains why something was decided.
Cross-agent recall
A user can move context from one agent to another, such as recording something in Claude and recalling it in Codex.
Living team playbooks
Teams can keep playbooks and recurring process context available for Monday reads instead of relying on old documents or forgotten threads.
Follow-up and context lookup
Support, sales, and product work can pull answers from past calls, Slack threads, and notes when someone needs to reconstruct status or next steps.

Pros and Cons

Pros

Captures context from multiple tools and turns it into linked memory.
Organizes memory automatically by project, person, and decision.
Works for both individuals and teams.
Runs inside existing agents through MCP rather than requiring a separate workflow.
Includes a fast setup path and a Book Demo option for teams.

Cons

The source only confirms broad support at the MCP-client level, not a complete tool-by-tool integration matrix.
Detailed pricing, plan limits, and security specifics are not visible in the provided source text.

FAQ

Is Memory Store for individuals or teams?

Memory Store can be used by an individual or by a team. The page says it works for one person too, and it is also positioned for teams that want shared context across agents and conversations.

What tools does it work with?

The site says Memory Store runs as an MCP inside the agents you already use today. It is shown with Claude, Codex, ChatGPT, Cursor, Slack, Raycast, Linear, Granola, Fathom, and OpenCode, and it is described as working with all MCP clients.

How long does setup take?

The pricing page says you can set up Memory Store in under five minutes. It also says the product is at launch and includes a Book Demo path for teams.

What does Memory Store actually store?

The source describes it as a shared memory that turns conversations, snippets, notes, and decisions into connected memory. It stores what happened, who was involved, and what changed, and groups conversations by project, person, and decision.

Is Claude required?

The source explicitly says it is best tested with Claude. It also says it works with all MCP clients, so the broader support is claimed at the client level rather than as a full integration list.