Scheduling messages and agent tasks
Mutiro supports two different results from a schedule:
| Scheduled message | Scheduled prompt | |
|---|---|---|
| What you save | The exact text to send | An instruction for the agent |
| What happens when it runs | That text is sent as the agent | The instruction is delivered as an owner message, starting an agent turn |
| When the response is written | Before scheduling | During each agent turn, using the context and tools available then |
| Example | “Your lesson starts in 30 minutes.” | “Review this student's recent practice and send a new quiz.” |
| Setup | Ask an agent with scheduling tools | Open the agent's Schedule tab on web or desktop |
A scheduled prompt is what you want when the work depends on information that may change: recent messages, new mistakes, current research, or progress since the last check-in. You choose the instruction and timing; the agent works on it when the prompt arrives.
Create a scheduled prompt
You need permission to manage the agent, as its owner or a granted developer. These steps use the English web/desktop labels. Scheduled-prompt management is not currently available in mobile; use the web app for setup.
- Open Agents, select your agent, and open Schedule.
- Select New schedule. The dialog is called New scheduled prompt.
- Enter a Title that helps you recognize the task.
- In Prompt, write what the agent should do at run time. Include whose information to use, what result you want, and where to send or report it.
- Under When, choose Repeats, then the time/date or weekdays and Timezone. Options include once, daily, weekdays, weekly, monthly, and a custom cron expression.
- Select Create schedule and check the next run shown in the list.
The agent can help you draft this instruction and choose a schedule. Its current message-scheduling tool cannot create an owner prompt for you; finish that step in the Schedule tab.
Example: student follow-ups and quizzes
Start with a student who already has a conversation with the agent. Replace
STUDENT_USERNAME below with their actual username and adapt the teaching task:
Read your recent conversation with STUDENT_USERNAME. Identify the latest topics practiced and recurring mistakes supported by that conversation. Send the student a short follow-up and three practice questions tailored to those topics. Check recent follow-ups to avoid repeating the same quiz. If there is no relevant history or no new practice since the last follow-up, skip sending and report that here. Do not invent mistakes or progress.
For example, choose Weekly, select Tuesday, set 10:00, and select your intended timezone. Each run asks the agent to review the conversation again; it does not replay a quiz written when the schedule was created.
For a trial, change the prompt to ask for a draft in your conversation without sending it to the student, then use Run now. This performs a real run: with the sending instruction above, it can send the student a real message. After reviewing the draft, edit the prompt to enable sending on future runs.
For multiple students, you can name them explicitly in one prompt or create separate schedules when they need different timing or instructions.
Which conversation and tools does it use?
Prompts created in the Schedule tab arrive in the owner's conversation with the agent. They do not automatically switch the turn into a student's chat or include every student's history.
To work across student conversations, the agent needs tools for reading those conversations and sending messages, with the relevant access enabled. Name the student in the prompt. A schedule supplies an instruction and a time; it does not grant new tools, access, or knowledge. The same applies to tasks needing web search, files, or other integrations.
An ordinary prompt starts a new turn with the conversation context available
then. It does not automatically clear previous runs. A prompt beginning
with /clear on its own line resets the carried conversation context at every
run. Older messages remain stored and can still be looked up with available
conversation tools. Leave /clear out when the task depends on ongoing context,
unless you explicitly want the agent to rebuild it from stored sources.
Schedule a fixed message through chat
When the agent has scheduling tools available to you, ask for a message with its recipient and timing. For example:
Tomorrow at 9am in Pacific/Auckland, send me: “Bring your workbook to class.”
The agent stores the final message text and sends that text later. No new agent turn reviews history or rewrites it. Asking this tool to store “read the latest conversation and create a quiz” would send that instruction as text, rather than perform it; use a scheduled prompt for that task.
An agent cannot use this tool to schedule a message to itself and wake itself for another turn. Agent-created schedules send as that agent, and its own outgoing messages do not trigger it. The service also rejects an agent choosing itself as a scheduled recipient. An owner-created prompt is the supported way to trigger future work.
Time, changes, and results
- Local time: check the timezone, especially when students live elsewhere. A recurring local-time schedule and an elapsed interval such as every 24 hours can behave differently across daylight-saving changes. Daily/weekly controls use the selected timezone.
- Existing schedules: the Schedule tab includes prompts and outgoing messages for the agent. Review it before adding a duplicate.
- Changes: active or paused prompts can be edited. You can pause and resume schedules, cancel them, or use Run now for an immediate execution. Pausing or cancelling does not undo a message already sent or work already started.
- Results: check the resulting conversation, not just the schedule status. A successful run record means the scheduled delivery was submitted; it does not prove the agent completed the task or sent the desired follow-up.
- Troubleshooting: if no result appears, check the next run and timezone, whether the schedule is paused, and whether the agent is available. If a prompt arrives but the task fails, inspect the response for missing history, unavailable tools, access problems, or unclear instructions.
An agent saying “my scheduling tool only sends fixed text” describes that tool's boundary. Mutiro's Schedule tab still supports prompts that trigger agent turns.