RMLRagMyLife
Local-first · Self-hosted · MCP

Local-first memory for your AI assistant.

From WhatsApp, email and documents. Runs on your machine. Your assistant — Claude today, any MCP client — answers with your own history as the source.

Source under PolyForm Shield 1.0.0, published at launch · Docker Compose · works on CPU only.

kitchen renovation quote
WhatsApp · Carpenter

“Updated quote attached: cabinets + countertop, delivery in 30 days.”

Email · PDF attachment

Commercial proposal — fitted kitchen, 3 finish options.

Meeting · Summary

Decided: option 2, paid in 3 installments.

Answered inside your assistant, through MCP — from an index that never left your machine.

One memory, working for you

Built for lives spread across a thousand conversations.

All of WhatsApp

Conversations, transcribed voice notes and images read as text — years of messages become text you can search.

Email and attachments

Every email and every attached PDF, spreadsheet or document, with the original file always one click away.

Documents and meetings

Document libraries, minutes and meeting transcripts organized by topic.

Search by meaning

Ask in your own words: “how much did I pay for the car repair?” finds the conversation, even without the exact words.

Gets better on its own

An evaluator measures search quality every night and the model is retrained on what you actually use.

People and loose ends

Who is who among your contacts, conversations waiting for a reply and topics still open.

Connects to Claude

A ready-made MCP server: Claude queries your memory right in the conversation, with the same login and the same permissions.

Yours, and only yours

Runs on your own machine. Your data stays home, and every access goes through login and per-area permission.

How it works

Three steps, all on hardware you control.

  1. Import

    A WhatsApp backup, your Gmail (through your own OAuth client) and folders of documents.

  2. Index locally

    Embeddings, transcription and OCR run on your machine. A GPU makes it faster; it is not required.

  3. Ask

    From the web panel, or from your assistant through the built-in MCP server.

Plans

One promise per plan — never both in the same sentence.

Cloud

Coming soon

We run the heavy parts; your assistant reaches your memory from anywhere.

  • Hosted inference: no GPU, no model downloads.
  • MCP relay for your assistant, wherever you are.
  • Encrypted backup of your index.
  • Price not set yet. The waitlist decides what ships first.

Only your email, language, interest and signup date are stored, only for this waitlist.

What leaves your machine

The honest version. One list per plan.

Self-hosted

One anonymous install event. Nothing else.

  • Embeddings, transcription and OCR run locally. No message, email or document is sent to a third party.
  • Cloud models (for image reading, for example) only if you opt in with your own API key, with a spending cap you set.
  • Anonymous install telemetry, used to develop the app: one event — version, OS, GPU yes/no, succeeded or failed. No identifier, no content, IP discarded. You are asked on first run; RML_TELEMETRY=0 turns it off. Paid plans have no telemetry. Every field will be listed in TELEMETRY.md (soon).
  • Live WhatsApp capture, if you install the plugin, runs on your own device.

Cloud (coming soon)

Listed, not hidden.

  • What you import — WhatsApp backup, mailbox, documents — is stored in your own tenant, encrypted at rest and isolated from other tenants.
  • Embeddings and transcription run on our servers instead of yours.
  • The MCP relay sees the questions your assistant asks.
  • Live WhatsApp capture still runs on your device; the Cloud only receives backups and exports.
  • The full list, terms and privacy policy will be published before the first paying customer.

Coming soon.

Join the list to hear when it opens.

Install guide soon — get notified