2pm.spaceAgents that show up
on time, every time
Tell an agent what to do and when — a sales summary at 8:00, a content digest every Monday, a follow-up sweep each night. It runs on its own, in your timezone, and files every answer where you can read it.
Put an agent on the clock
Pick the agent, set the time, read the results.
Pick an agent and a job
Choose the agent and write what it should be asked, one message per line — each line is sent and answered as its own item.
Set when it runs
One-time, daily, weekly on the days you tick, monthly on a date, or a cron expression — in the timezone you choose.
Read what it did
Every run lands in the task's history with each answer, its tokens and its cost. Run it now, pause it or cancel it from the list.
Pick an agent and a job
Choose the agent and write what it should be asked, one message per line — each line is sent and answered as its own item.
Set when it runs
One-time, daily, weekly on the days you tick, monthly on a date, or a cron expression — in the timezone you choose.
Read what it did
Every run lands in the task's history with each answer, its tokens and its cost. Run it now, pause it or cancel it from the list.
A schedule that knows it's running an agent
Not a cron job firing a webhook — a run with a conversation, tools and a log.
Five kinds of schedule
One-time on a date, daily at a time, weekly on chosen days, monthly on a day of the month, or any cron expression. Every schedule keeps to the timezone you set, not the server's.
Results go where they're needed
The agent runs with its tools, so the report can go out as it is written: a Telegram message to your sales group, an email, a push to your operators, a row in a Drive table.
See the tool catalogHalf the cost, when it can wait
Switch a simple agent to Batch API and its runs go through the provider's batch queue — about 50% cheaper, finished within 24 hours. For Anthropic and OpenAI models.
Memory across runs, if you want it
Start a new conversation each run, keep adding to the latest one so the agent remembers last time, or post into a specific conversation.
It stops when you say
Set an expiry date or a maximum number of runs, and choose how many messages run side by side — from one to ten.
Every run on the record
Each run keeps its status, its start and finish, and every message's answer or error with tokens and cost — also collected under Settings › Logs.
Every scheduling option
What the Create Task dialog offers, field by field.
Schedule
- One-time, at a date and time
- Daily at a time
- Weekly, on any days you tick
- Monthly, on a day of the month
- A custom cron expression
- Any timezone
Execution
- Immediately, 1–10 messages at a time
- Batch API, about 50% cheaper, within 24 h
- Batch for simple agents on Anthropic or OpenAI models
- Agents with a Human Review step can't be scheduled
Input & conversation
- One message per line, each its own item
- A new conversation every run
- Or continue the latest conversation
- Or a specific conversation by ID
Limits
- An expiry date and time
- A maximum number of runs
- Runs counted in the list, e.g. 6/12
Control
- Run now
- Pause and resume
- Cancel
- Edit or delete
- Search, and filter by status or mode
History & logs
- The status of every run and every message
- Each answer or error, in full
- Tokens and cost per message and per run
- A Scheduler Runs tab in the workspace logs
Cron and scripts vs. the Scheduler
The same morning report, two ways.
A cron job, a script and a server to keep alive — to ask one question.
Pick the agent, write the question, set the time.
A run fails at 3 a.m. and you find out a week later.
Each run's status and error sit in the task's history.
Every report starts from scratch.
Continue the latest conversation, and the agent remembers the last one.
Paying full price for work nobody needs until tomorrow.
Batch API runs cost about half.
Scheduler questions
What people ask before they automate their first report.
Let the agent keep the schedule
Build it once, put it on the clock, and read the results over coffee.