Always in sync. Never start over.

Teach it once.
It never forgets again.

Claude, Cursor, and Copilot all start from zero — same project, same questions, every new chat.

Persistly gives them one shared memory that lives in your GitHub repo: real commits, a running history of every decision, read automatically before every reply.

No credit card required
Free to start Your code never leaves your GitHub repo Connected in under 2 minutes
persistly.map — change history
reading before reply
a3f9c2e2 min agopersisted
+ theme.js — decision: CSS variables for dark mode
~ settings.jsx — added toggle, matched ProfilePage style
Read by Claude Desktop before responding. No prompt needed.
f11a08dyesterdaypersisted
+ api/rateLimit.js — added, tied to existing auth middleware
- hardcoded 100ms retry delay — replaced with backoff
Same memory, read by Cursor in a different session.

Persistly connects over MCP (Model Context Protocol), the open standard that lets AI tools read and act on outside data. One memory, shared across every tool you switch between.

Not your tool's built-in memory. Cursor and Copilot only remember inside their own app. Persistly's memory lives in your repo, as real commits — so it's still there when you switch tools, machines, or AI models.
Works with
Claude Desktop Claude Code Cursor VS Code Copilot Windsurf Goose
Disconnected, until Persistly sits in the middle
Same three tools. Watch what changes.
Without Persistly With Persistly
Every tool starts from zero. One shared memory, always in sync.
Persistly
Sitting idle Reads context first
What's the context?
Got it — full context
Claude
Claude Full context, automatically
What should I do?
Right where I left off
Cursor
Cursor Same memory, new session
No memory of this project
Every change, logged
GitHub
No memory of past decisions Every decision, a commit
Step 1
Create a Space
Connect a repo and describe your project once. Persistly writes a persistly.context file into it — your AI's permanent memory, updated as the project evolves.
Step 2
Connect your tools
One config paste into Claude Desktop or Cursor. The AI gets tools to read, write, and update memory — the same config works for VS Code Copilot and Windsurf too.
Step 3
Never explain twice
Every reply starts with the full context, read automatically. Every file it writes is a real commit, logged in persistly.map.
Questions
How is this different from my AI tool's built-in memory?

Built-in memory lives inside that one tool. Persistly's memory lives in your GitHub repo as real commits, so it's still there when you switch from Claude to Cursor, switch machines, or hit a usage limit and open a different AI entirely.

Does my code ever leave GitHub?

No. Persistly reads and writes directly to the repo you connect. There's no separate copy of your code sitting in a Persistly database — the repo is the source of truth.

What happens if I stop using Persistly?

Nothing disappears. Everything Persistly writes is a normal commit in your own repo, so your project history stays exactly where it is, in your GitHub account, whether or not you keep using Persistly.

Do I need to change how I code?

No. You keep working in Claude, Cursor, or whichever tool you already use. Persistly runs in the background over MCP — it reads context before the AI replies and logs changes as it works.

Is it free?

Yes, free while in beta. If paid tiers come later, existing usage patterns won't be pulled out from under you without notice.

Free while in beta. Open source. Built by one developer who got tired of repeating himself.

Free to use · Open source · Built for developers who ship with AI every day
Spaces
Persistly

Welcome to Persistly

Create a Space and connect your GitHub repo. AI will read and write your files automatically — with full project memory.

📋 Overview
🗺️ Persistly Map
🔌 Connect AI
Space Info
Name—
GitHub Repo—
Brand—
Industry—
Tone—
cortex.context View on GitHub ↗
Loading…
cortex.map auto-updated by AI
🔗 Your Personal MCP URL
This URL connects Claude Desktop or Cursor directly to your Space. It's unique to you — no downloads, no local setup.
Generating your URL…
🖥️ Claude Desktop Config
Replace the entire contents of your Claude Desktop config file with this. Requires Node.js installed (nodejs.org).
📁 Config file:
Windows: %APPDATA%\Claude\claude_desktop_config.json
Mac: ~/Library/Application Support/Claude/claude_desktop_config.json
Loading…
!
Fully quit Claude Desktop (system tray → right click → Quit) then reopen. Look for 🔨 hammer icon.
✓
That's it — just talk normally. Claude reads your project automatically and saves changes to GitHub as it works.
⚡ Cursor Config
Create or open ~/.cursor/mcp.json and paste this:
Loading…
🤖 Connect Other AI Tools
Claude Desktop and Cursor are set up above. Click any tool below for its exact setup steps.
🪟 VS Code (GitHub Copilot)
VS Code uses a different config key (servers, not mcpServers) and requires Agent mode.
1
Create .vscode/mcp.json in your project, or run MCP: Open User Configuration from the Command Palette for a global setup.
2
Paste this:
Loading…
3
Open Copilot Chat → switch mode to Agent → click the 🛠️ tools icon to confirm Persistly tools are listed.
🌊 Windsurf
1
Open ~/.codeium/windsurf/mcp_config.json (Mac/Linux) or %USERPROFILE%\.codeium\windsurf\mcp_config.json (Windows). Use the gear icon in the Cascade panel → MCP Servers if you'd rather not find it manually.
2
Paste this:
Loading…
3
Click Refresh Servers in the Cascade MCP panel.
⌨️ Claude Code (CLI)
1
Run this in your terminal (inside your project folder):
Loading…
2
Start a session with claude as usual — Persistly tools will be available.
🪿 Goose
1
Run goose configure → Add Extension → Command-line Extension.
2
When asked for the command, paste:
Loading…
💬 ChatGPT
ChatGPT connects to remote MCP servers through its own UI rather than a config file, and currently expects either no auth or full OAuth — not a custom header — so the setup below is best-effort and may not expose write access the same way Claude/Cursor do. Requires a paid plan (Plus/Pro/Business) with Developer Mode enabled.
1
Settings → Connectors → enable Developer Mode if not already on.
2
Add custom connector → paste your MCP URL (shown at the top of this page) as the server URL. Leave auth as "none" — token-based header auth isn't supported in ChatGPT's connector UI yet.
3
In a new chat, click + → More → Developer Mode → select Persistly as a source.
⚠️ Claude.ai (web) does not yet support adding custom remote MCP servers as of this writing — only Claude Desktop and Claude Code do.
🛠️ Tools Available to AI
read_context()
Called automatically first — reads all files, cortex.context, and change history
save_progress(path, content, is_new)
One call saves the file to GitHub and updates the change history — no separate map call needed
ask_codebase(question)
Ask a targeted question — routed through Groq, returns a short answer instead of full file dumps to save tokens
read_file(path)
Read one specific file from GitHub
update_context(content)
Updates cortex.context when project goals or stack change
list_files()
Lists all files in the GitHub repo
delete_file(path)
Deletes a file from GitHub
⚡ Claude is told to call read_context automatically when it connects — you no longer need to ask it to.
1
Details
2
GitHub
3
Repository
Create a Space
A Space is a persistent AI memory for one project or client.
🏢 Brand Context (optional)