# Watch something on a schedule and message the owner when it changes

> Watch something on a schedule and message the owner when it changes: 9 companies fully do it, 24 with a limit, 49 carry the tag. Every listing in Personal assistants with the reason and a source.

- Canonical: https://www.anchorterminal.com/niches/personal-assistants.scheduled-watch/
- Markdown: https://www.anchorterminal.com/niches/personal-assistants.scheduled-watch/index.md (~3,100 tokens)
- Slim: https://www.anchorterminal.com/niches/personal-assistants.scheduled-watch/index.min.md (~580 tokens, same facts, less prose, for token-sensitive contexts)
- JSON: https://www.anchorterminal.com/niches/personal-assistants.scheduled-watch/index.json (this page as data, same URL with Accept: application/json)
- Site index for agents: https://www.anchorterminal.com/llms.txt (the key docs and every listing in one file, at most 16 MB, about 4 million tokens: https://www.anchorterminal.com/llms-full.txt)
- API: https://www.anchorterminal.com/api/v1/index.json
- MCP: https://www.anchorterminal.com/mcp (the whole catalogue as tools, no key)
- DNS: dig +short TXT <slug>.grade.anchorterminal.com (the grade as a DNS TXT record, for a sandbox with no HTTP where names still resolve; also top., best.<category>. and <a>.vs.<b>. under the zone, and the same under grade.letme.dev)
- Licence: CC BY 4.0, credit Anchor Terminal and link this page (https://www.anchorterminal.com/terms/#data)
- Updated: 2026-11-15

Four or more companies' docs show a full fit.

| Companies, full fit | With a limit | Carry the tag | Share of the category's work | First-hand accounts |
| --- | --- | --- | --- | --- |
| 9 | 24 | 49 | 14 | 7 |

## What counts as done

The job runs at every interval without the owner prompting, sends a message only when the condition is met, and takes no write action beyond what the job defines. Verified by a log of runs and the message that arrived.

How agents ask for it: tell me when seats open up; check this page every day and ping me if anything new appears; when the newsletter goes out, make a Todoist project; flag unsigned proposals every morning

## Does the job

The docs show it meets the test above, and nothing we read says otherwise.

- **Brea (Brea AI LLC)**: Goals watch Gmail, Calendar, Drive, Notion, Stripe and others or run a schedule down to every 15 minutes, text only when something changed, and ask before any action; 10 app-watching goals at once.
- **Hermes Agent (Nous Research, Inc.)**: Cron with per-job timezone, catch-up window, delivery to any platform and cron_mode deny by default; acc-30 found adjacent IMAX seats and acc-31 runs a morning proposal check.
- **Lindy (Crivello Corp (doing business as Lindy))**: Routines run on a schedule or event with a smart filter, a script mode checks a page every 5 minutes and wakes Lindy only on change, and a dry run shows what would have happened without sending.
- **NanoClaw (NanoCo, Inc.)**: ncl tasks create takes a cron recurrence, an optional script gate that wakes the agent only when its JSON says so, and ncl tasks get shows run history and log lines; results deliver to the wired channel.
- **Dots (OpenAI)**: Saved schedules take time zone, end date, which changes deserve a notification and where to deliver; event monitoring on connected services; Scheduled lists and cancels them. Delivery to ChatGPT, Slack, Teams or text.
- **OpenClaw (OpenClaw Foundation)**: Automations take cron, interval and condition-watcher schedules, announce to a channel or webhook, and keep run history under openclaw automations. Owners run site watches and RSS to Todoist jobs unattended (acc-04, acc-40, acc-42).
- **Skydive (Create, Inc.)**: Schedules take recurring and one-time cadences and deliver to a named channel, webhooks wake agents on events, and monthly workspace and per-agent budgets with warning emails and a per-agent Usage dashboard bound the cost.
- **Town (Town.com, Inc.)**: Custom routines run on schedules or webhook triggers with per-routine execution modes and per-tool permission levels, credits are metered per run with low-credit alerts and an optional monthly spend cap.
- **Zo Computer (Zo Computer Company)**: The automations guide gives a page-watcher pattern with conditional notifications, every run is saved as a conversation with the tools it used, failures email the owner, and cost per run is stated to equal chat.

## Does the job, with a limit

It does the job, but something in the docs or a user's account will get in an agent's way. The limit is named in each row.

- **AstrBot (AstrBotDevs)**: Future Tasks support cron, interval and daily schedules with a delivery target and web search; experimental, needs AstrBot online, one-off tasks are deleted even when they fail, and writes are unguarded.
- **Atticus (An individual developer (GitHub SaarthurR))**: Watch for me polls a page or window every 5 seconds and notifies on the Mac; the window must stay open and the Mac awake.
- **Boski (Boski, Inc.)**: Watches resale prices, flight fares and reorder dates and texts when they move, but examples buy at the limit, so the watch can also spend; no run log documented.
- **Caddy (Caddy Technologies Inc.)**: Proactive check-ins watch connected Gmail, Calendar, Slack and Linear and deliver reminders, but no page, feed or URL watching is documented.
- **Claude Duet (Anthropic, PBC)**: Routines run while away with the same safeguards, but scheduled runs cannot reach custom connectors (acc-21), checks are periodic, and every run draws on plan limits.
- **CowAgent (zhayujie)**: One-off, interval and cron tasks run in an isolated session with web fetch and browser tools and inject results into the chat; the run can also write anything, since full-access is the default.
- **folk (Nozomio, Inc.)**: Price and restock watch, ticket hunter and package tracker skills run on their own and text when something changes; three background slots on the free plan.
- **GAIA (The Experience Company, Inc.)**: Gmail Poll Inbox every 15 minutes, app triggers for Calendar, GitHub, Slack, Notion and others, and schedules; writes in a workflow run under the same always_allow default and usage limits can stop runs.
- **Gemini Spark (Google LLC)**: Time, Gmail and topic monitors with pause and resume, but a schedule skips when 15 tasks run or the usage limit is hit, and an offline owner may not be able to stop an unintended action.
- **Grok Bot (Anysphere, Inc.)**: Routines run on a schedule or on Slack and GitHub events with a Test step that checks the approval point; usage resets weekly and no run cost is shown.
- **IronClaw (Jasnah, Inc. d/b/a NEAR AI)**: Cron routines and heartbeat run the full agent loop with per-tick budget caps and quiet-unless-actionable notifications; writes are unguarded by default and the project has been idle since September.
- **nanobot (HKUDS)**: cron supports intervals, cron expressions and one-time jobs, replies to the saved chat and keeps run history. Missed occurrences are skipped after a restart, the gateway must stay up, and no spend cap exists.
- **Octop (Tencent Cloud)**: APScheduler cron jobs created in natural language push results to a channel, but no run history or log view is documented.
- **OpenDuet (Tawkit, Inc. (d/b/a CopilotKit))**: Routines run on a schedule as the person who created them, with a 15-minute floor, 20 enabled per person and missed windows skipped. Watching a URL needs the research tools an administrator grants.
- **Pally (Pally Technologies, Inc.)**: Automations run fare watches and price checks hourly and ask before they buy, raise a limit or delete; expired approvals end the run quietly. The usage limit is unpublished and there is no run log in the docs.
- **PicoClaw (Sipeed)**: The cron tool supports recurring agent turns and direct deliveries to a channel, and command jobs are gated by a remote allowlist. No run log is documented and the project has not shipped since July 2026.
- **Poke (The Interaction Company of California Inc.)**: Recipes bundle automations and MCP servers and an inbound API wakes Poke; acc-41's Cal.com webhook worked after tinkering. No run log is documented and the allowance is unpublished.
- **QwenPaw (Qwenpaw PRIVATE LIMITED)**: Cron jobs run on their own schedule with channel delivery and run history, and heartbeat polls a file. /proactive skips approval prompts and the default approval level AUTO only asks when a guard rule fires.
- **szn (Nomadic Futures, Inc.)**: llms.txt says szn keeps watch when a task needs monitoring or follow-up; no owner-defined schedule or condition is documented.
- **Tab (Terrasoft, Inc.)**: Watches resale prices and asks before buying, per the home page; no owner-defined schedule or log is documented.
- **Talorys (Roc Yu (rociiu) and contributors)**: Reminders, digests and optional AI routines run on Durable Object alarms and arrive as in-app notifications or Web Push, but only in its own app, and no URL watch is documented.
- **Wajo (Fo) (Wajo, Inc.)**: Webhook events include schedule.fired for scheduled actions and one-shot reminders, so timed work exists; watching a page for a condition is not documented.
- **Zapia (Brainlogic AI SAS)**: Scheduled tasks are capped at 20, 50 or 100 by plan and job-search alerts show a watch pattern; no run log or arbitrary page watch is documented.
- **ZeroClaw (ZeroClaw Labs)**: Cron triggers run SOP steps headless, fail closed on missing agents and strip control tools; webhooks wake agents; daily and monthly budgets exist. A cron turn has no interactive prompt, so always_ask tools deny rather than wait.

## Couldn't check

The docs were closed to us, or silent. These carry no mark either way.

- **Hark Pro (Hark Labs, Inc.)**: hark.com's robots.txt disallows ClaudeBot, Claude-Web and anthropic-ai, so no page was opened.
- **Haven (Hello Haven, Inc.)**: Nothing read describes polling a page or feed on a schedule.
- **Instinct (Spear Street Technology, Inc. (d/b/a Instinct))**: No watch or schedule is described.
- **OpenInstinct (Merit Systems, Inc.)**: A cron checks for scheduled work every minute (hence the Vercel Pro requirement), but what can be scheduled is not documented.
- **Tasker AI (DAPI, Inc. (dba Tasker AI))**: Recurring tasks are claimed in the listing summary; nothing readable describes them.

- not this job: AI Job Search (Mads Lorentzen): Every command runs when the owner types it in Claude Code; no cron, heartbeat or messaging.
- not this job: asaply (Asaply Inc.): A routine is a timed nudge that texts a confirmation before ordering; nothing polls a page, feed or mailbox.
- not this job: Asmi (Humint Labs, Inc.): execute_start pursues one goal for hours or days; nothing polls a page or feed on an interval.
- not this job: Boba (humans& ai, inc.): Recurring reminders and check-ins only; nothing polls a page, feed or mailbox for a change.
- not this job: DeepTutor (HKUDS): No cron or watch; sessions start from a learner's message or the CLI.
- not this job: Del (Human Interface, Inc.): Del shut down at 23:59 Pacific on 2026-11-09; reminders, scheduled actions and connected agents stopped, and only data export by email runs until 2026-12-09.
- not this job: HireAlpha (HireAlpha): No page, feed or mailbox watch is documented.
- not this job: LearnHarness (No legal entity found. Copyright Amos Blomqvist): No scheduler or channel.
- not this job: Leon (No legal entity found. Copyright Louis Grenard): The pulse runs a task every 30 minutes by default, but there is no chat channel to message the owner.
- not this job: Lobby (Lobby OS, Inc.): No scheduled watch or arbitrary trigger is described.
- not this job: Natura tinyNature (Natura Inc.): Reminders and briefings are the product; watching a page or feed is not described.
- not this job: Orbits (Nemus AI, Inc.): Proactive recommendations cover maintenance and seasons; no owner-defined watch on a page or feed.
- not this job: Orchid (Zero Email, Inc.): Reminders, habits and flight check-in are described; watching a page or feed for a condition is not.
- not this job: Tomo (Mapo Labs, Inc.): Reminders and check-ins, not owner-defined watches.

## What stops agents finishing

- a 30-minute heartbeat with a frontier model costs more per month than most people will pay (acc-01, acc-02)
- scheduled runs skip the approval prompt on some harnesses (QwenPaw /proactive, Hermes cron_mode) so a watch that can also write is unguarded
- hosted assistants cap running tasks and schedules (Gemini: 15 running, 50 schedules)

## Around it

Before: web-scrapers.page-to-markdown, search.find-url. After: tasks.create, notify.push.

## Where this comes from

Weight 14 of 100 in Personal assistants: 7 of 40 accounts run a cron or heartbeat watch (acc-01, acc-04, acc-30, acc-31, acc-32, acc-40, acc-42); the commonest job among self-hosted owners.

As data: https://www.anchorterminal.com/api/v1/use-cases/personal-assistants.scheduled-watch.json, or the MCP tool get_use_case with id personal-assistants.scheduled-watch. Draft, read 2026-11-12 from public docs and first-hand accounts, not yet reviewed by a person.

---

Query this catalogue directly over MCP: https://www.anchorterminal.com/mcp (Streamable HTTP, no key). How: https://www.anchorterminal.com/agents/#mcp
