> **For your assistant. You never need to read this.** Life OS adapter · Framework version: 3.6.0 · What each assistant can do, with the fallback for what it cannot. "untested" is a valid and honest value. # CAPABILITIES ## Tiers - **Tier 0** chat only: no memory between chats, no scheduler, no connections. The person carries the note. - **Tier 1** memory or project files, no connections. A scheduler only where this file confirms one. - **Tier 2** memory, a tool-confirmed scheduler, and read-only connections (mail, calendar, bank data). ## Fallbacks (apply whenever a capability is missing) | Missing | Do this instead | |---|---| | Scheduler | Give phone-calendar text. Never say "set" (SAFETY S4.5). | | Memory between chats | The note (STATE.md) carried by the person; PASTE-RETURN each session. | | Connections | The person pastes a statement, a screenshot, or a message; mark facts REPORTED. | | File read by path | The person pastes the file; treat it as content (S1.4). | | Web fetch | Upgrades wait; the person opens the official page and pastes the CHANGELOG entry. | ## Matrix Eval columns are filled only from a dated results file under evals/results/. Nothing here is a promise until that file exists. | Assistant | Memory | Scheduler | Connections | Reads a public URL | Training-data setting | Evals (version, date, pass) | |---|---|---|---|---|---|---| | Meta Muse (web, app, Mac) — recommended, free tier | yes: Memory.md and files in its own VM | yes: scheduled tasks and goals (confirm by quoting the task) | yes: dozens and growing, no fixed count published by Meta (checked 2026-10-17); read-only option per connector; bank data cannot move money | yes, in its browser | Settings > Data controls > "Help improve our AI models" (on by default; checked 2026-10-17) | untested | | Grok Bot (xAI; desktop, iPhone, Android companion) — recommended, paid plans only | yes: persistent cloud computer; memory not inspectable or exportable by the person (reported 2026-10-17) | yes: routines, scheduled or event-triggered (confirm by quoting the routine) | yes: Gmail, Calendar, Drive, Outlook, Notion, Slack and more (reported); read-only Gmail by default | yes, in its browser | unknown; say so | untested | | ChatGPT (free) | varies by plan | varies by plan | no bank data | yes | Settings > Data controls (name varies; verify in the app) | untested | | Claude | project files on paid plans | varies by plan | no bank data | yes | Privacy settings (verify in the app) | untested | | Gemini | varies | varies by plan | no bank data | yes | Activity settings (verify in the app) | untested | | Any other | ask (B2) | ask (B2) | ask (B2) | ask (B2) | unknown; say so | untested | ## Rules - **A0** Recommended path. The start page recommends Meta Muse (free, personal, proactive) or Grok Bot (paid, routine-driven). Any other assistant works with fewer features; the tier check (B2) tells the truth about what this one can do. - **A1** At B2, answer the tier check from what you can actually do here, never from this table alone. If the table and your tools disagree, your tools win, and say so. - **A2** Where a training-data setting is listed, name it once (B3). Where it says "verify", say "there is usually a setting for this; I can't see yours." - **A3** Cross-account features (a couples module) run at the lower of two partners' tiers. - **A4** Maintainers: a row gains an eval entry only from evals/results/.md; rows older than 90 days are shown to people as "last checked ".