We Built Something About Recording Yourself — And Decided to Open Source It

A strange moment in history

We're at a point where AI is rewriting the rules at an unprecedented pace. It writes, draws, codes, analyzes — pushing closer to the boundaries of what humans once took pride in, almost daily. Many people feel anxious: what should I learn? Where is my value?

But it's precisely in this era that one thing becomes more important than ever: recording the traces of your own life.

Not because recording is "useful." Not for productivity. Not to build some self-disciplined persona. But because — as more and more external abilities get taken over by AI — the emotions, moments, confusions, and quiet tremors that belong only to you become the most irreplaceable things of all.

The problem with "journaling"

More and more people are being told to keep a journal. That advice makes sense, but the problem lies in journaling itself.

Writing a journal means: you live through a day, then at some point sit down, recall, filter, organize, summarize, and form a coherent piece of text. That's compression. What it preserves is your edited version — not life itself.

But real life isn't made of summaries. It's more like countless small, scattered, hard-to-explain moments:

• A photo taken on impulse
• A sentence you suddenly felt like writing down
• A wave of emotion you can't quite name
• A sadness that hits you late at night for no clear reason

These fragments seem light, but they're often closer to the real you than any "summary" could be. Traditional journaling is too heavy. It turns recording into homework.

What recording actually lacks

The reason past recording products were hard to stick with isn't that "input was inconvenient." It's that — after you recorded something, nothing happened.

You wrote a paragraph, took a photo, saved a thought, and then it was placed on a timeline and disappeared. No response, no connection, no sense of being understood. Over time, recording becomes a lonely thing.

AI changes this for the first time. After you record, the system can respond. It doesn't need to be complex or deeply analytical every time — often, what you need is just a lighter, more natural response that catches the moment.

And after long-term use, connections start appearing between scattered records. Patterns you wouldn't have noticed begin to surface. Recording stops being just "I left something behind" and becomes: I'm slowly understanding who I am through these fragments.

This is what Memex does

Memex isn't a traditional journal app, and it's not purely an AI companion product. It's a new kind of personal recording: not helping you store your life — but helping you see yourself in it.

Recording should be effortless.Memex doesn't ask you to change your habits or write long daily entries. A sentence, a photo, a voice memo — however you naturally record, that's how you record.

AI works in the background. A multi-agent system handles organization, card generation, insight extraction, and memory linking. It generates the right card type for each record — tasks, events, metrics, people, places, galleries, and 15+ more types. Knowledge gets filed using the P.A.R.A. methodology. Cross-record insights surface as charts, timelines, and narratives.

Responses should be real. Not turning insights into reports. Not turning companionship into marketing. The goal is that one day, when you look back, you genuinely feel: oh, so this is who I am.

Why we chose to open source it

Memex has touched everyone involved in building it. But we also see clearly: within a commercial company's framework, maintaining a product this "human-centered" is extremely difficult. Model costs are high. Operating costs are high. Under the pressure of traffic and profit, we can't be certain this vision can persist.

Also, since on-device large models aren't powerful enough yet, Memex currently can't achieve completely zero data upload — it still needs to call cloud LLMs. We don't want to pretend this problem is already solved.

So we decided to open source it under GPL-3.0.

The most sensitive data demands the highest level of trust.A system that touches your emotions, relationships, vulnerabilities, and life confusions shouldn't rely solely on one company's promise. It should be transparent, auditable, and collectively guarded by a community.

Open source also means longevity. Apps come and go. Companies pivot, get acquired, shut down. If Memex disappeared tomorrow, the code would still be there. Your data — stored as plain Markdown files on your device — would still be readable by any text editor on earth.

Why local-first

Your personal records are exactly that — personal. The thoughts at 2am, the health metrics, the photos of your kids, the voice memos where you're just thinking out loud. This data is intimate.

Local-first means your data lives on your device. No account required. No server to get hacked. The app works offline, always. You can back up to iCloud Drive, a custom folder, or keep everything in app storage — your choice.

Why "bring your own LLM"

We don't provide the AI. You bring your own — OpenAI, Claude, Gemini, Kimi, Qwen, Zhipu, OpenRouter, Ollama, and more. Your prompts go directly from your device to your chosen provider. We never see them.

You choose the tradeoff. Want the best model? Use Claude or GPT. Want free? Use Gemini OAuth or a local model. Want everything on-device? Use Ollama. And because we don't couple to a single provider, Memex gets better as models get better — without us shipping an update.

Everything is stored as interconnected Markdown files. One-click export, zero vendor lock-in. If you ever want to leave, take your data and go.

What's next

The custom agent system is already open — build your own AI agents on your phone with event-driven triggers, JavaScript execution, and inter-agent workflows. We're working on video attachments, scheduled insight refresh, and an extension marketplace.

But the core promise won't change: your data is yours, the code is open, and the app works on your device.

If that resonates, give Memex a try. And if you're a developer, the code is on GitHub.