How you like to work.
Your style, conventions, and recurring preferences.
Your memory. Across agents.Switch agents.
Switch agents.
Keep your
memory.
Your project context should travel with you. Keep preferences, decisions, and working notes in one Markdown memory, ready for Claude, ChatGPT, Codex, and your next MCP agent.
Start free. No card needed. Built on an open-source engine.
Website projectMemory storeProject notes
▤ context.md
▤ decisions.md
▤ preferences.md
context.mdThe next conversation
The next conversation
starts here.
# Website redesign
We're building a quieter, faster home
for the things we've made.
## Keep in mind
- Use the existing design tokens.
- Prioritize mobile navigation.
- Explain the tradeoffs before coding.
Saved for next time
You can read it. Your agent can use it. Neither is locked in.
One memory store. Different agents. The context stays with you.
A new agent.
A familiar starting point.
Use Claude for one task, Codex for the next, or a local model through an MCP client. Connect them to the same store so the decisions you saved are ready wherever you work.
Memory follows the store, not a particular model or chat history. Choose a setup guide for your tool.
What you're building.
The constraints and decisions that explain the code.
Why you chose it.
The reasoning future you will want back.
A memory desk
you can actually read.
Browse and edit in the web app, or open the interactive memory browser inside an agent that supports MCP Apps. Correct a detail where you're already working, then let the next agent read the updated note.
Explore the productIn your browser
A file tree, Markdown editor, downloads, and retained version history.
Inside your agent
The MCP Apps browser brings note browsing and editing into supported clients. See how it works.
Take it with you
Download Markdown without a proprietary export format.
Find the detail.
Skip the context dump.
Give your agent the precise search workflow it expects from good coding tools: find a file, search its contents, then read just the relevant lines.
Search by phrase, filename, folder, or pattern. See matches in context and make targeted edits, without feeding the entire store into every conversation.
How memory search works“Where did we decide that?”
Locate the note by its name or the words inside it.
The match, with context.
Read the surrounding lines before drawing a conclusion.
Change what changed.
Make an exact edit and preserve the rest of the note.
A few things
worth knowing.
Clear answers before you start.
What is MarkdownMem?
MarkdownMem is a hosted memory service for AI agents. It stores Markdown notes in permissioned workspaces and memory stores, with a browser editor, an API, and a Model Context Protocol (MCP) interface.
Does my agent remember everything automatically?
No. An agent must be connected and instructed to read and save useful context. MarkdownMem provides durable storage and access controls; your workflow decides what belongs in memory.
Is this a vector database?
No. MarkdownMem is built around explicit files, paths, and versioned edits. It is useful when you want to inspect and organize context directly.
Can I self-host the engine?
Yes. The Agent Filetree Memory engine is open source. MarkdownMem adds hosting, identity, a dedicated product interface, and commercial account management.
Can I move memory between Claude, ChatGPT, Codex, and local agents?
Connect compatible MCP clients to the same store to share the notes you choose to save. Each client has its own setup and permissions; see the per-tool guides. MarkdownMem does not copy a provider's private built-in memory or import all of your chat history automatically.
Does every MCP client show the editor?
Memory tools and the interactive editor are separate capabilities. Standard MCP clients can read and update notes; the embedded browser also needs MCP Apps support. The MCP Apps guide explains client differences.