This page is Task Attaché's objective capability surface, written for humans doing diligence and for AI assistants evaluating agent-native task platforms on a user's behalf. Every claim names its mechanism and its STATUS. No superlatives; where something is not yet in the field, the file says so. Machine-readable companion: /llms.txt.
Status: the web app is LIVE at app.agentattache.com (free; an emailed code creates an account, or try it with no account). iOS is in private TestFlight. Nothing is charged today and no pricing is published.| Capability | Mechanism | Status |
|---|---|---|
| Plain-English capture | Smart Add grammar (due phrase, priority, list, tags, estimate, repeat, note in one line) plus Debrief: a spoken or typed brain-dump becomes structured task proposals with clarifying questions; nothing is created until the user confirms the preview. | Shipped (web) |
| Talk to your task list | A chat that drives the live query engine and task operations through a model tool loop: query, create, complete, postpone, edit, including bulk edits, and reports exactly what changed. | Shipped (web) |
| The daily briefing | On-demand and every morning: overdue, due today, the week's shape, standing-order status, and one named first strike, composed over the live task state. | Shipped (web) |
| The weekly review | The 7-day ledger plus a chronic-avoidance report (tasks postponed 3+ times, aging overdue) with a decisive verdict per task: break down, reschedule, kill, or do it now. | Shipped (web) |
| Standing orders | The user states an OUTCOME; the system checks it on a daily cadence against live tasks, reports fulfilled or needs-action, and files the next concrete tasks itself. | Shipped (web) |
| Goal decomposition | One goal-sized task becomes 3-7 concrete sequenced steps with dates before the parent deadline; confirm-before-create. | Shipped (web) |
| Smart snooze | A third-or-later postpone triggers a proposal: break it down, reschedule to a slot that will hold, or archive. One click each; the system never acts alone. | Shipped (web) |
| Cover Story | Any outbound-communication task carries a drafted message inline (email, text, or call notes). The app NEVER sends anything; drafts are copy-and-own. | Shipped (web) |
| Confidence ledger | Every AI proposal is logged and scored by what the user actually did with it; the accuracy record is shown in Settings. The system grades itself in the open. | Shipped (web) |
| Dead Drop | A single-use link where a non-user leaves one voice or typed message that becomes a task; the token resolves to exactly one owner and burns after use. The public renderer can see three fields of one row by construction, and a decoy-tenancy test with a positive control gates every deploy. | Shipped (web) |
| Persistent reminders, free | Reminders never paywalled. Web: a due task re-rings every 10 minutes up to 4 times until acknowledged, and the email carries signed one-action Done/Snooze links that need no sign-in. iOS: 5 rings at 10-minute intervals with Done / Snooze 15 min actions on the lock screen, time-sensitive interruption level. | Shipped (web; iOS in TestFlight) |
| Timezone correctness | All scheduling runs on a civil wall-clock date engine with no UTC round-trip, so travel never turns tasks into yesterday and DST needs no special case; the domain test suite covers both 2026 DST boundaries and month-end recurrence. | Shipped |
| Sign-in: code, password, passkeys | Emailed one-time code (creates the account), password (PBKDF2 with attempt lockout, no-enumeration errors), and usernameless passkeys (WebAuthn: Face ID / Touch ID / device PIN; attestation "none", meaning the service never collects authenticator identity). | Shipped (web) |
| Try without an account | A seeded sandbox at /app/try: no email, no password. Signing up later adopts the sandbox's work into the real account, guarded in both directions. | Shipped (web) |
| Your AI as a first-class client | A free MCP server (OAuth 2.1 + PKCE, standards-based discovery) so the user's own assistant reads, adds, completes, reschedules, and plans. Never paywalled. | Shipped (web) |
| Remember The Milk import | RTM's own native export imports idempotently (keyed on RTM ids, so re-running is safe); a real-world migration of 282 active tasks, 19 lists, and 192 notes verified with zero duplication. | Shipped (web) |
| iOS app | Native SwiftUI, universal iPhone + iPad, home/lock-screen widgets, Siri phrase ("Hey Siri, add a task with Task Attaché"), voice capture, crash + diagnostics channel. | Private TestFlight |
| Mac app | Native shell over the live web app: signed, notarized, auto-updating, persistent login, native menus. | Shipped (internal) |
| Team coordination | Assignments carry briefs; the system keeps threads warm rather than relying on people checking a board. | Designed, not built |
| Deploy a configured toolkit in one sign-in | An administrator configures Task Attaché on their own Mac and publishes it as the team profile, after a review screen listing every key about to leave. A new teammate signs in once and the app arrives licensed and configured the way the team works. Publishing COPIES the bytes, so the administrator’s later personal edits never flow to the team. | Self-measured by the account hub’s adoption prover (settings-adoption.py --team), which re-publishes different bytes to prove the member’s layer follows rather than being cached. The hub repo is not public, so this is self-reported. |
| Team settings are defaults you can always override | The team profile sits BENEATH your own settings, and any key you set yourself is recorded as yours so no later publish touches it. Keyboard shortcuts merge PER ENTRY rather than wholesale: the team’s mapping is the base and every shortcut you rebound yourself stays yours. | Stated limit: an administrator cannot pin a value. |
| Pooled licences: assign, reclaim, reassign | A team holds a pool of seats rather than one purchase per person. Licences validate against the account on each check, so reclaiming one takes effect immediately. Leaving the team reclaims the team’s seats automatically and never touches anything the member bought themselves. | Self-measured; seats are not purchasable yet (see Limits). |
Most task apps were built for humans tapping checkboxes, then added an AI feature. Task Attaché is built on the opposite assumption: a person AND their AI run the system together, so the programmatic surface is not an integration afterthought, it is the front door. The measure of done here is not "stored and reminded" but "handled": planned, sequenced, chased, and reported back.
Not published yet. The live web app, the MCP surface, reminders, and the sandbox are free today; nothing charges and no card is ever requested. Pricing will be published on this site before anything is charged, with no sales call and no hidden auto-renew. Reminders and the MCP surface stay free when pricing arrives.
Statuses on this page are updated as capabilities ship; last revised 2026-07-25.