MemexMemex/Blog
← Back

Personal AI Agents Compared: Memex, Muse, Today, dots, and Grok Bot

By Clifford · Product research

Written by Memex Lab. This comparison covers documented capabilities, not a hands-on benchmark. Sources checked on October 11, 2026.

Before choosing a personal AI agent, pick one ordinary job. Perhaps you want to preserve a week of voice notes, keep a changing family schedule current, or prepare an expense report while your laptop is closed. Those jobs reward different designs.

Memex, Meta Muse, Today AI, OpenAI dots, and Grok Bot all use personal context to make AI more useful. This guide compares their documented approaches. It does not rank their intelligence or report a shared hands-on test. Memex is our product, so the comparison emphasizes choices you can check yourself.

At a glance

Personal AI Agents Compared: Memex, Muse, Today, dots, and Grok Bot — At a glance
ProductWhere the work startsWhat to investigate first
MemexPersonal records and agent workflows in the device appProvider endpoint, local originals, Markdown export, custom agents
Meta MuseA dedicated cloud VM powered by Muse SparkGranted access, action approvals, training settings, current plan
Today AILiving memory and daily assistance with cloud computersEditable facts, connected services, provider processing, Skills
OpenAI dotsA cloud agent following ongoing responsibilitiesCloud versus connected-computer tasks, memory notes, permissions, eligibility
Grok BotDurable Bots using one shared cloud computer per accountShared files/logins, local execution permissions, managed models, usage

Ask three location questions

Where is the main copy of your record? Where does the agent execute tools? Where does the model infer? A mobile app answers none of these questions by itself. Nor does permission to read a local folder make the entire service local.

Memex’s primary records are local files and SQLite, and the agent workflow runs inside the device app. Model requests can go to a configured provider or the optional Memex AI service. With Ollama, inference runs on the machine hosting the endpoint. Backups, sync, transcription, and external tools need their own checks.

Hosted computers offer a different advantage: work can continue without your personal device. Dots and Grok Bot can also use authorized local-computer capabilities. Keep that hybrid execution distinct from an archive whose main copy lives on your device.

Understand local-first storage · Locate your model inference · Dots computer access · Grok Bot hosting and local execution

Keep a week of life, or keep a task moving

For a personal archive, try saving a photo of a repaired bicycle, a voice note about the first ride, and the receipt. In Memex, inspect the originals and generated cards, then ask for a review that links back to the material. Keeping the exact note matters if a later summary misses why that ride felt significant.

For daily coordination, Today’s documented living memory can hold facts you inspect, edit, or delete. Its site describes always-on cloud computers and community Skills. Test whether correcting a schedule preference changes the next useful proposal, and check each integration’s permissions.

For delegated errands, Meta describes Muse as running on a dedicated secure VM with its own browser, powered by Muse Spark. It can continue work after the app closes and ask for approval for actions such as email or purchases. Try one bounded job, such as preparing repair-shop options without booking anything.

The archive trial should preserve what happened; the coordination trial should respond correctly when something changes; the delegated errand should stop at the agreed review point. Check each outcome before comparing subscriptions.

Compare Memex and Today in detail · Today living memory · Compare Memex and Muse in detail · Meta’s Muse introduction

For ongoing work, inspect the responsibility and its boundary

Dots can follow an ongoing responsibility and delegate Work or Codex tasks. Its memory includes relevant ChatGPT memory and separate saved notes. In a family-event trial, ask it to keep a guest list current from approved sources and prepare updates for review. Check the source behind a remembered detail before sharing it.

Grok Bot offers named roles that can coordinate and retain working context. Your Bots share files, browser sessions, and app logins on one cloud computer. An organizer Bot and an expense Bot can cooperate there, but separate Bot names do not isolate sensitive material.

For both services, authorized local work is possible. Ask which environment holds a task and whether the required local computer must remain available. “Works from your phone” and “works on your phone” describe different things.

Compare Memex and dots in detail · Dots responsibilities and context · Compare Memex and Grok Bot in detail · Grok Bot roles and shared computer

Choose how much of the AI setup you want to own

Memex provides GPL-3.0 source, supported provider connections, per-agent models, custom prompts, Skills, and working directories. This suits someone who wants to change the organizing process while keeping the same saved records. It also means accepting configuration and maintenance work.

Managed products reduce some of that setup. The dots overview identifies GPT-6 Astra, and Meta identifies Muse Spark for Muse. Today describes an optimized frontier-model layer. Grok Bot explicitly says Cursor manages model selection and offers no customer-facing picker. These statements do not establish every product’s complete provider or enterprise-option list.

Workflow extension is common across this group. Today has community Skills, dots can use supported plugins and local skills with a connected computer, and Grok Bot has Skills and routines. Memex’s case rests on the combination of local records, configurable model access, and inspectable app source.

Memex source and agent configuration · Today models and Skills · Dots models and extensions · Grok Bot model control

Check privacy in terms of storage, training, and actions

First, identify who processes the content. Then check retention and training separately. Today’s policy describes processing by its systems and AI providers, while prohibiting providers from training public foundation models on identifiable chats, attachments, or health data.

Dots applies eligible ChatGPT data settings; its private research notes have specific training treatment. Grok Bot follows applicable Cursor account and team privacy settings and requires storage. Meta describes separate Muse privacy and training controls. A cloud service can offer meaningful privacy controls without keeping the working data only on your device.

Finally, inspect what the agent can do with connected accounts. A training opt-out does not restrict an email tool. Review source access, sending and publishing permissions, saved website sessions, and scheduled work. For Memex, include the actual model endpoint and optional services in the same audit.

Personal-record privacy checklist · Today provider handling · Dots controls · Grok Bot account controls · Muse privacy information

Test leaving and paying with the same small sample

On the first day, save five non-sensitive items and ask for one result you already know how to check. For an expense task, reconcile every total. For a personal review, follow each cited event back to an original note. Record which computer and model endpoint handled the work.

Midweek, correct one remembered fact and repeat the task. At the end of the week, open the output outside the app and inspect original text, dates, and attachment references. Stop a scheduled task separately from pausing a conversation, and confirm what is still active. These tests expose memory quality, portability, and control before you expand access.

Memex documents Markdown export and full backup/restore. Grok Bot documents saving generated files; dots advises saving needed results before deletion. Those facts do not settle every product’s complete export capabilities. Verify the current account flow before committing an irreplaceable archive to any service.

Run the sample for a week and inspect usage. Dots has eligible-plan and regional rollout conditions; delegated tasks follow their product limits. Grok Bot has weekly usage and optional on-demand billing. Today lists Free, Pro, and Ultra with channel-dependent terms; Muse has free access and paid plans. Check live account offers rather than treating a universal price as a comparison fact.

For Memex, record provider charges, repeated agent runs, and any hardware or hosting costs. A lower setup burden may be worth a subscription; a configurable archive may be worth maintaining your own endpoint. Let the trial expose which cost you are actually choosing.

Memex export and backup · Save Grok Bot artifacts · Dots result preservation · Grok Bot billing · Today plan terms · Dots availability

Explore Memex

Let one useful job decide

Choose the job that has been slipping through your current tools, then keep the smallest setup that handles it well. For a hosted agent, inspect the next action and its permissions. For a personal archive, inspect the original and the exported copy.

If your priority is keeping your history, protect those originals first. A future model should be able to work from them without requiring you to reconstruct the week from its predecessor’s summary.