“What are the agents running on Thursday?”
Work for agents sits in a backlog with no days on it. Nobody can say what runs when, or what lands on the same morning.
A ConvOps week and month board for what your scheduled AI agents execute, sized to what your workspace actually finishes. Spread work before its deadlines, plan next week in one pass, and let each item start itself on its day.
capacity = points done per week, last 4 weeks · no history: unknown, never 0
to plan · 2 items with no day
ConvOps Plan is a week and month board where a team decides what its agents execute and when, sized to workspace capacity, and can start work on its planned day.
ConvOps is the operations layer for AI agents. It holds your team's process as workflows with gates, a shared memory, scheduled and unattended runs, programs, a plan board, metrics and a full audit trail. It does not sell or host models.
key facts · october 2026
best for
not for
updated
Work for agents waits in a backlog with no days on it. The overload shows up on the deadline, and a schedule that drifted an hour leaves no trace.
“What are the agents running on Thursday?”
Work for agents sits in a backlog with no days on it. Nobody can say what runs when, or what lands on the same morning.
“Will all of this fit before the deadline?”
The plan assumes a week the team has never finished. The overrun shows up on the deadline, not on the Monday before it.
“Why did the 08:00 run start at 07:00?”
A schedule kept in UTC drifts an hour when the clocks change, and a run that could not start leaves no trace.
Days by lanes, grouped by project, by executor or flat. Work with no day waits in the tray until you drag it onto one.
Each article sits on its publish day in a lane per publication, and starts itself at 07:00 on that day.
week board · lanes by project · auto-run
Reconciliations, accruals and variance notes are dated up to close day. The month counts each day, and slipped items are flagged before the close.
month view · planned days · slipped flags
Lanes by executor show what each executor carries each day, and the recurring client reports sit on the same days as chips.
lanes by executor · schedule chips
Forty migration batches wait in the tray with no day. The tray shows the true count, and each batch gets a day when someone places it or a spread does.
tray · Now and Next · place or spread
Drag a fix onto Thursday and it moves to Now. Take it off again and its auto-run is cancelled with it.
placement · Now cap · unplan cancels auto-run
Work that comes in phases fits the board as it is: the ConvOps implementation plan template turns an approved plan into one child task per phase, and each phase task can go on its own day.
Capacity is what your workspace completed per week over the last four weeks. Spread work onto the least-loaded days before each deadline, and see the overrun before you commit.
6 items to spread · size first
A dry-run spread places the forty migration batches on the least-loaded days before cutover, shows each day's points and warns when a week goes over capacity, before anything is committed.
spread · dry run · week over capacity
A hundred and twenty document reviews across legal, financial and IT lanes, with over capacity set to refuse. An overbooked spread is refused whole, so no workstream is half planned.
lanes by project · over capacity: deny · all or nothing
With no finished weeks yet, capacity reads unknown instead of zero, so the first plan is not told it has no room. Weekend days count at a quarter.
capacity · unknown, never 0 · weekend factor
Review what slipped, choose against capacity, confirm. Three steps and one commit, and the dry run says what it found before anything is written.
Each Friday the audit lead plans next week's evidence collection: slipped requests first, picks against the meter, a dry run with spread order set to deadline first, one commit. Each item then starts at 06:00 Europe/Berlin on its day.
Plan next week · capacity meter · deadline-first spread · auto-run
Sales operations picks next week's account research against capacity, so every brief has a day before the review it feeds.
Plan next week · deadlines · spread
Give a planned item a start time and an executor. On its planned day, at that time in the task's time zone, ConvOps dispatches it.
Repeat work every few minutes, hourly, daily, on weekdays, weekly, monthly, or with a repeat rule. Each schedule has its own time zone and a preview of its next runs.
example · treasury · a market-open briefing per desk
08:00 local = 23:00 UTC, the day before
08:00 local = 12:00 UTC, 13:00 from 1 Nov
next 5 runs
recent
A monthly schedule on day 1 at 08:00 Europe/London creates a fresh report task each month, so every month keeps its own record.
monthly · day 1 · new task per run
A custom repeat rule wakes the same vendor-review task every quarter, so its notes and runs stay on one record. A daily schedule creates a new sanctions-list check each morning.
custom RRULE · wake · daily new task
Three weekday schedules at 08:00 in Singapore, Dublin and Chicago run one triage workflow, each on its own local morning.
weekdays · a zone per schedule
A once schedule starts the cutover at 02:00. If it has not started 15 minutes later, the Schedules page flags it overdue.
once · overdue after 15 minutes
Capacity, the spread, auto-run and schedules are plain rules over workspace settings, readable from the app, REST and an AI client.
Capacity is the average points completed per week over the configured weeks. Points come from size, unsized items count as M, and a week is 5.5 units: five weekdays and two weekend days at a quarter each. It is one figure for the whole workspace.
plan.capacity_weeks 4
plan.points_s|m|l 1 / 3 / 8
plan.unsized_as M
plan.week_units 5.5
plan.weekend_factor 0.25Each item goes to the least-loaded working day on or before its deadline, in the configured order. A dry run returns where each item would land without writing. If any item cannot be placed, the whole spread is refused. The over-capacity setting decides whether an overbooked spread is allowed, warned about or refused.
POST /api/v1/plan/spread dry run or commit
plan_manage commit_spread the same from an AI client
plan.spread_order size_first | deadline_first
plan.spread_weekends false
plan.spread_over_capacity allow | warn | denyA planned day is a date. Auto-run adds a time, and ConvOps dispatches the task at that day and time in the task's time zone through one deferred dispatch. The executor is named explicitly and custom-runner executors are refused. A task already finished by then is not dispatched, and the reason is auto_run_stale.
PUT /api/v1/plan/auto-run/{task_id}
DELETE /api/v1/plan/auto-run/{task_id}
plan_manage set_auto_run | clear_auto_runA cadence kind is described back in plain English, a custom RRULE is shown as written, and runs are expanded in the schedule's IANA zone, so daylight-saving gaps and overlaps are handled. A rule with no future run is refused at create. A refused run creates nothing and writes one blocked activity per occurrence. Run now does not advance the cadence.
schedules_read get | summary | upcoming | runs | preview
schedule_create | schedule_attach | schedule_update
schedule_list | schedule_delete | schedule_trigger
POST /api/v1/schedules/preview next 5 runs, UTC and localPlan is the ConvOps week and month board where a team decides what its agents execute and when. Tasks with no day wait in a tray until someone places them on a day. Capacity comes from what the workspace completed over the last four weeks, a spread fills the least-loaded days before each deadline, and a planned item can start itself at a set time on its day, on the executor you choose.
Open Schedules in ConvOps and create a schedule: choose whether each run creates a new task or wakes an existing one, then set the cadence, the time zone and the executor. Cadences run from every few minutes to monthly, or a custom repeat rule. ConvOps previews the next five runs before you save. For one-off work, place the task on a day on the Plan board and turn on auto-run with a start time.
ConvOps Plan is for teams that hand recurring or deadline-bound work to AI agents and need to decide what runs on which day: operations, finance, compliance, content, research and engineering teams. It fits when the work is bigger than one chat, such as a quarter-end close, an audit evidence list or a content calendar, and when the team wants to see a full week before committing to it.
ConvOps measures capacity from what your workspace completed: the average points done per week over the last four weeks. Sizes S, M and L count 1, 3 and 8 points, unsized items count as M, and weekend days count at a quarter of a weekday. Capacity is workspace-wide. With no history yet it says unknown, never zero, so a new team is not told it has no room.
Yes. An AI client connected to ConvOps can create, update, list and trigger schedules with the schedule tools, and read previews and upcoming runs with schedules_read. The same rules apply as in the app: every schedule has a cadence and its own time zone, and a rule with no future run is refused. Inside an isolated run, creating or changing schedules stays off unless that environment grants it.
ConvOps refuses the run, logs it with the reason and keeps the schedule enabled. A run is refused when its project is archived or unknown, its program is inactive, its executor is missing, or its auto-run is stale. No run is created, one blocked activity is written for each occurrence, and the Schedules page shows the issue with a link to the fix.
Yes. Every ConvOps schedule has its own IANA time zone, UTC by default, and its runs keep local time across daylight-saving changes. A weekly 07:00 run in Chicago stays at 07:00 local when the clocks change, and only its UTC time moves, from 12:00 to 13:00. The preview shows the next five runs in local time and in UTC before you save.
With ConvOps, any work your agents already run through a workflow can repeat on a schedule: a daily reconciliation check, a weekly compliance evidence pull, a monthly report on day 1, a morning briefing per region. Each run follows the same workflow steps as a manual one, approval gates included, and a run that cannot start is logged with its reason instead of vanishing.