Memex vs Today AI: Personal Memory, Privacy, and Customization
By Clifford · Product research
Written by Memex Lab. This comparison covers documented capabilities, not a hands-on benchmark. Sources checked on October 10, 2026.
A useful personal AI needs more than your next instruction. It needs context: a promise you made, a project that stalled, or the reason you changed a plan. Memex and Today AI both work with that kind of continuity.
Today emphasizes a living memory and help throughout the day. Memex starts with personal records on your device and lets you choose the models and agent workflows around them. The distinction matters most when you want to keep more than the facts an assistant needs for its next task.
At a glance
| What matters | Memex | Today AI |
|---|---|---|
| Starting point | Text, photos, and voice saved as personal records | Living memory and proactive daily help |
| Agent execution | In the device app; inference follows your model configuration | Always-on cloud computers |
| Memory and records | Local source records, cards, knowledge files, and agent memory | Inspectable, editable memory facts |
| Extending workflows | Custom agents, prompts, Skills, and GPL-3.0 app source | Community Skills |
| Cost structure | Model and infrastructure costs follow your chosen setup | Free, Pro, and Ultra plans; check your purchase channel |
Useful memory is not the whole record
Today describes memory facts that you can inspect, edit, and delete. Memex is therefore not the only option for people who want to correct what their assistant remembers.
Consider a note saying: 'I moved Thursday's run because dinner with my sister mattered more this week.' A calendar needs the new time. An assistant may need the updated commitment. Your personal record may also need the reason, the original wording, and the photograph you attached.
Memex keeps source records alongside generated cards and knowledge files. That gives you material to revisit when your interpretation changes. A concise memory can help an agent act; an archive can help you remember what a week actually felt like. Decide which of those you want to preserve.
Today's explanation of living memory · How Memex organizes journal records
A cloud assistant and a device-local workflow
Today's site describes always-on cloud computers and community Skills. This makes it a relevant comparison for Memex, rather than simply a chatbot with a calendar.
Memex's journal workspace and agent workflow live in the device app. The primary records are local files and SQLite data. AI inference is a separate choice: cloud providers receive the context needed for a request, and an Ollama endpoint runs inference wherever that endpoint is hosted.
The practical trade-off is availability. A local workflow is subject to phone battery, OS background limits, and access to its model endpoint. Do not assume that Memex will execute tasks continuously while your device is unavailable. A cloud service also has service and connectivity dependencies, but a different operating model.
Today's capabilities · Understanding the primary copy of your records
Privacy depends on the path your context takes
Today's policy describes processing on its systems and through AI providers, while prohibiting those providers from training public models on identifiable chat content, attachments, or health data. Read that distinction before equating cloud processing with training.
Memex offers a different starting point: a local journal archive and configurable model access. With a directly configured provider, requests go to that provider; the optional Memex AI service proxies model calls. Neither mode justifies a blanket claim that all content stays on the phone.
If you want to reduce external processing, review the actual endpoint, any sync or backup destination, transcription settings, and tools that call other services. Open source makes the implementation inspectable. It does not automatically make every configuration private or every dependency local.
Today privacy policy, AI provider section · Memex model connections · A privacy checklist for personal records
Customize the way your records become useful
Start in Memex's Model Configuration, available from the avatar menu. You can connect providers such as OpenAI, Claude, or a supported Ollama endpoint, and assign models to individual agents. Then adjust a system prompt, add a Skill, or scope an agent's working directory. The GPL-3.0 app source lets you inspect, build, and modify the implementation under that license.
A practical experiment is a reading log. Save a passage and your response, then configure an agent to file the idea under a project you care about. A second agent could prepare a review of related records. Start by running and checking a small example; custom automation still needs configuration and evaluation.
Today also supports community Skills. In Memex, try changing the model for that reading agent while leaving the saved passage in place. The record and the workflow around it can evolve separately. You gain control over that setup, and take responsibility for maintaining it.
Memex custom agents and Skills · Choosing a local model setup
Compare costs with your own week of use
Today lists Free, Pro, and Ultra plans, with availability and pricing varying by purchase channel and other conditions. Check the current offer for your account instead of relying on a universal monthly price.
For Memex, account for the provider, model, amount of context, and number of agent runs you choose. A photo-heavy workflow or repeated long reviews can have different costs from short text notes. Local inference may avoid per-request cloud charges but still requires hardware and an available endpoint.
Use a small, non-sensitive example in each product. Check what was saved, what you can correct, which actions need your approval, and how you would retain the original material if you stopped using the app. Compare that outcome before comparing a subscription with an estimated API bill.
Start with what you want the AI to remember
If you mostly need help keeping the next day coordinated, a managed daily assistant is a sensible starting point. If you want the notes, images, reasons, and changes behind that day to remain a personal archive you can shape, Memex is worth trying.
For a comparison focused on delegated errands and cloud execution, continue with Memex vs Meta Muse below.