Claim your memory.

How to install

In your chosen LLM, open Settings and follow this flow:

→ Connectors → + Add Connector → Custom MCP Connector

1. Name it "Persistent Flow" Copy
2. Paste the link https://flow.persistentflow.com/mcp Copy  → Add Connector
3. → Sign in to Persistent Flow
4. Repeat for every LLM you use, and enjoy your context following you everywhere.

Basic commands

• Demo: say "can you start the Flow demo" for a guided walkthrough.
• Onboarding: tell your AI "set up my flow" for a quick guided setup.
• Menu: say "open my Persistent Flow menu" to see what Flow can do.

Specific steps per LLM

Claudepaid plan

Custom connectors need a paid Claude plan (Pro / Max / Team / Enterprise). Works in Claude.ai and Claude Desktop.

1. Settings → Connectors → Add custom connector.
2. Name: "Persistent Flow". URL: https://flow.persistentflow.com/mcp.
3. Connect → sign in on the Persistent Flow login page.

ChatGPTinstall in browser; works in app too

Setup has to be done on chatgpt.com in a browser: ChatGPT's app management isn't in the mobile/desktop apps. Once installed, it works everywhere: browser and apps. Needs Plus / Pro / Business.

1. Open chatgpt.com → Settings → Apps → Advanced settings → Create an app.
2. In the dialog: Name Persistent Flow · Linking Server-URL + paste https://flow.persistentflow.com/mcp · Authentication OAuth · tick the “I understand and want to continue” risk box. Icon is optional (any PNG ≤ 10 KB, e.g. favicon-black.png from /press).
3. Click Create → sign in on the Persistent Flow login page.

Mistral Vibe (formerly Le Chat)free plan works

Custom MCP connectors work on Vibe Free as well as paid plans. Le Chat was renamed Vibe on 28 May 2026 (same URL chat.mistral.ai, same login, plans and history carried over); its modes are Vibe Chat, Vibe Work, and Vibe Code.

1. Connectors → + Add Connector → Custom MCP Connector.
2. Name: "Persistent Flow". URL: https://flow.persistentflow.com/mcp.
3. Connect → sign in on the Persistent Flow login page.

Local AI clientsGoose · Continue · Cline · Cursor · Zed

Persistent Flow is a standard remote MCP server, so any MCP-capable client connects. Goose, Continue, Cline, Cursor, and Zed are good options. llama.cpp's web UI (the chat interface served by llama-server) now ships its own MCP client, so it connects directly, no third-party wrapper needed. Ollama and LM Studio work too.

No API key needed. Flow issues no keys. Sign-in is OAuth via WorkOS: a browser login, nothing to paste. The catch: the client has to run the OAuth handshake. Claude, ChatGPT, and Mistral do this automatically; many local clients do not. A client that cannot sign in itself bridges through mcp-remote, a small local proxy that runs the OAuth flow for it (needs Node). Point your client at this as a stdio command: npx -y mcp-remote https://flow.persistentflow.com/mcp. It opens your browser once to sign in, caches the token, and presents Flow to your client. For JSON-config clients, set the command to npx and the args to -y mcp-remote https://flow.persistentflow.com/mcp. Still stuck? [email protected].

llama.cpp specifics: the MCP client lives in the web UI, not in the llama-server process itself. Browser-to-server calls need the CORS proxy: start llama-server with --webui-mcp-proxy (spelled --ui-mcp-proxy on newer builds) and enable the “use llama-server proxy” toggle on the MCP server entry in the UI. Both halves are required.

1. Install your local AI client of choice (Goose, Continue, Cline, Cursor, Zed, …).
2. Add Persistent Flow as a remote MCP server: URL https://flow.persistentflow.com/mcp, transport streamable-HTTP.
3. Sign in via WorkOS when prompted, same login = same Flow as your Claude / ChatGPT / Mistral. If your client cannot sign in, see the auth note above.

Each client's exact dialog differs (some take JSON in a config file, others have a UI). Search "MCP server" in your client's docs and use the URL above.

… and many more.

Stop re-explaining yourself

Every new AI chat starts cold. You explain your business, your customers, your team, your decisions, and your documents again.

Persistent Flow holds that context once, so any connected AI can use it.

One memory across every AI

Save something in Claude. Look it up in ChatGPT. Ask Mistral to reason about it.

Same Flow. Every chat. No copy-pasting between assistants.

Build your Soul

Beyond facts, Flow can hold you, your worldview, opinions, voice, and the calls you'd defend. Say "build my soul" and any AI you connect can think and write as you, not as a generic assistant. Refresh it anytime as your Flow grows.

What goes into your Flow?

• Your Business: what you do, who you serve, and where you're going.
• Your Knowledge: documents, notes, research, training, and lessons learned.
• Your Decisions: the choices you've made and the reasons behind them.
• Your Market: customers, competitors, and industry insights.
• Your Projects: what you're building, tracking, and improving.

Privacy that scales

Choose the right level of privacy for every piece of information:

• Standard for your working knowledge.
• Private for sensitive data, protected by your own passphrase.
• Vault for information you never want an AI to see, off-LLM unlock; stronger isolation options on the roadmap.

No new app to learn

Use the AI tools you already love. Persistent Flow simply gives them memory.

Power users: type terminal-style commands like /flow menu, /flow soul, or /flow find <X> to jump straight in.

What's coming next

• Shared Workspaces: work with co-founders, teams, and organizations in a shared Flow while keeping personal records private.
• Greater Data Control: choose where your data is stored and how it is protected as we expand our infrastructure options.
• More AI Clients: Claude, ChatGPT, Mistral, and local AI clients (Goose, Cursor, Cline…) work today. Gemini and others will follow as their MCP support broadens.
• Web & Mobile Access: browse, search, and manage your Flow through dedicated web and mobile apps.
• Smarter Search: semantic search, improved retrieval, automatic summaries, and better discovery across your knowledge.
• Secure AI for Sensitive Information: we're exploring ways to process highly sensitive information with stronger privacy guarantees and more controlled AI environments.

Memory → Privacy → Security → Roadmap

Help shape the roadmap

Tell us what matters most during onboarding. Your feedback directly influences what we build next.

Do I need an API key or credentials?

No. Flow uses OAuth sign-in (WorkOS). You add the MCP URL, click connect, and sign in in your browser. Same login in every AI = one Flow.

I removed the connector or cleared my login. Did I lose my data?

No. Your records live on our servers, tied to your login, not to the connector or any local token. Removing a connector or clearing your device credentials only cuts the connection. Reconnect with the same login and everything reappears. Use the same account you started with: a different one opens a separate empty Flow, and your original data stays safe under the original login. If a client shows "invalid credentials" on reconnect, remove it and add it again so it does a clean sign-in.

Where does my data live?

Persistent Flow is built in Sweden, but today runs on US-based infrastructure.

Your records are stored in a managed database, while authentication is handled by a separate provider. We chose this setup to move quickly and validate the product.

We're transparent about this: today, your data is hosted on US-controlled infrastructure.

How is my data protected?

Your data is encrypted in transit and at rest.

Every Flow is private to its owner, and we cannot browse user records. We only see aggregate usage statistics and the feedback you explicitly choose to share.

You can choose between three privacy levels:

• Standard: available to your connected AI assistants.
• Private: protected by your personal passphrase.
• Vault: off-LLM unlock; never sent to any AI. Stronger isolation and secure-AI processing options are part of the roadmap.

What your AI sees

Persistent Flow doesn't push data anywhere. When Claude / ChatGPT / Mistral asks Flow for a record, Flow returns it to that AI as part of the conversation. What each AI does with that data afterwards is governed by their policies (Anthropic / OpenAI / Mistral), separate companies with separate privacy terms.

What is the long-term vision?

Our goal is a local-first architecture.

Over time, your memory should live primarily on your own devices, with optional encrypted backups to storage you control.

Persistent Flow becomes the layer that synchronises, indexes, and retrieves your knowledge, rather than the place where it lives.

We're not there yet, but that's where we're heading.

Where we're going

The current version of Persistent Flow is hosted in the cloud because it was the fastest way to validate the idea and learn what users actually need.

As the platform grows, our vision is to provide greater control over where data is stored and how it is protected.

Over time, we want users and businesses to be able to store their data in the region that makes the most sense for them, whether that's Sweden, Europe, the United States, or their own private infrastructure.

Security and privacy are not features we'll add later. They are fundamental design principles. Our goal is to build a system that gives users more control over their data, reduces unnecessary exposure, and aligns with the strongest privacy and security standards available.

We're also exploring ways to handle highly sensitive information without exposing it to general-purpose AI models. This includes investigating secure processing options for Vault records, trusted AI environments, and alternative ways of accessing sensitive information while maintaining stronger privacy guarantees.

The same applies to ingestion. We're actively researching how highly sensitive information can be added to Persistent Flow in a way that gives users maximum control over what is shared, processed, or retained.

Today, we're focused on proving the value of Persistent Flow. As the platform grows, we'll continue investing in infrastructure, privacy, and security while keeping our core mission the same: giving you one trusted memory layer that works across all your AI tools.

We're not there yet, and we want to be honest about that.