How to build a knowledge base from your messages

The short answer: your messages already are a knowledge base

A knowledge base from messages is a searchable, askable record built from the conversations you already have. You ask a question in plain words, and it answers with the exact text your answer came from.

Think of it as a personal knowledge base you don't maintain. It is a private store of what matters to you, without the filing.

It is for people with wide circles and no system for keeping up. It hands back the recommendations, links, promises, and dates that already live in your texts.

What a knowledge base is, and what makes one "personal"

A knowledge base is a searchable store of information you can look things up in. Most of the ones you meet online belong to companies, holding help articles for their support teams.

If you searched the phrase online, most results describe that company version, built for support agents. The personal kind barely shows up.

A personal knowledge base holds your life instead of a company's. A reference definition of what a personal knowledge base is calls it an electronic tool an individual uses to capture and later retrieve personal knowledge.

That same definition notes it differs from a traditional database because it holds subjective material particular to the owner. Your version is yours: the plans and the small details only you would care about.

Why the best stuff ends up buried in your messages

The problem starts with volume. According to CTIA's annual wireless survey, Americans exchanged nearly 2.2 trillion SMS and MMS text messages in 2024, more than any year besides 2020.

None of that was saved on purpose. The good stuff arrives in the middle of a conversation. It is a restaurant a friend swears by, or a link you meant to read.

Picture a link a colleague sent you last quarter, buried a thousand messages back. You know it exists. Finding it by scrolling is the part that fails.

Memory does not hold it for long either. In PLOS ONE, a 2015 replication study confirmed the Ebbinghaus forgetting curve, so memory decays substantially in the hours and days after learning without review.

We still want all of it within reach. In a 2016 Pew Research survey of 1,520 U.S. adults, 77% said they like having so much information at their fingertips. That was a decade ago, and the volume has only grown.

Why the usual fix (a note-taking system) quietly fails

The common answer is a note-taking app or a "second brain." Stanford's knowledge management framework describes personal knowledge management (PKM) as five steps: capture, categorize, organize, retrieve, and share. That is the work of organizing what you know, done by hand.

A Massachusetts General Hospital library guide calls those practices people use to manage knowledge strategies to collect, organize, process, and use knowledge for better decisions.

The trouble is upkeep. A busy month happens, the entries stop, and the tool quietly goes out of date. A knowledge base you stopped feeding is worse than none, because you trust it and it is wrong.

Good note-taking tools exist, and they still need you to show up every day. The discipline is the whole cost, and it is the first thing a busy week takes.

Here is the split in plain terms.

How a knowledge base from messages actually works

Automatic capture is the fix. A knowledge base from messages reads your existing Messages history and turns it into something you can search and ask. Every answer comes with its source.

It runs on-device, which means the work happens on your own Mac rather than a company's server. There is no contact to type and no tag to pick.

You do not learn a query language. You type the way you would text a friend, like "What hotel did Dana book?"

When Sanctuary answers, it shows the message the answer came from. You can trust it or scroll to the rest of the thread yourself.

So the recommendation a friend sent in March is findable in seconds, with the original message attached. You already wrote it down. It just was not anywhere you could ask.

What you can actually do with it

Once your messages are searchable, a few everyday jobs get easier. Each one runs on the same underlying memory of your people.

None of these need a new habit. They read the same conversations you were already having, so the work happens without you.

• Find recommendations: ask who suggested that ramen place and read the exact text back.
• Follow up on promises: see the intro you said you would make and the reply you still owe.
• Build a reading list: turn the links people text you into a read-it-later reading list.
• Remember people: keep a personal CRM from your Messages without typing a single entry.

A quick example

Here is an illustrative scene. The names are made up.

Last spring, your friend Maya texted you about a small Italian place in Cincinnati she loved. Six months later you are headed there for work and cannot recall the name.

You ask your knowledge base, "What restaurant did Maya recommend in Cincinnati?" It answers with the name and the original message, dated last April.

Why it should stay private and on your Mac

The record of who you are should not live on someone else's server. That is why Sanctuary keeps your library on your own Mac.

Local-first means the library stays on your machine: no account, no sync, no telemetry, and nothing to log in to. You can read how Sanctuary handles your data before you install anything.

Do not take our word for it. Turn off your wifi, and Sanctuary still answers. It keeps working on a plane or a weak connection, since nothing depends on a server answering.

When you want a bigger model, cloud AI runs on a key you provide, inside hardware your Mac verifies first. It is off by default. You bring the key, so the bill and the access stay in your name.

I built Sanctuary because I kept losing these threads myself. A friend once texted me a doctor's name, and a year later I could not find it when it mattered.

Ravi, founder

Connecting your knowledge base to an AI assistant

You can also connect it to Claude. Sanctuary includes an MCP server, and MCP (Model Context Protocol) is a standard way for an AI assistant to query another app.

This is the part a general assistant cannot do on its own. It has read the whole internet and none of your life.

For people already running Claude Desktop or Claude Code, this turns a general model into one that can answer questions about your own life.

Behind switches you control, your assistant can ask Sanctuary about your people. You decide what crosses: only what it learned, or the full words of your messages.

How to build your own, step by step

Getting started takes four steps, and none of them ask you to file anything.

1. Install Sanctuary on an Apple Silicon Mac.

2. Give it permission to read your Messages history, and let it build your library.

3. Ask questions in plain words, like who recommended a place or whose birthday is next.

4. Connect Claude if you want your assistant to use that context.

The first build can take a little while if you have years of history. After that, new messages fold in as they arrive.

You never file a note or tag a contact. The capture is automatic, so the library stays right even in a busy month.