“Can I let it run while I sleep?”
Most tools make you choose: watch every move in a chat, or hand it to a script and hope.
The same ConvOps workflow runs in your AI client or unattended on a schedule. You decide which checkpoints it has, if any: an approval, a required metric, a report, or none. Same workflow, same rules, with or without you watching.
attended by default · schedules · work pools · a record for every run
Yes. The same ConvOps workflow runs attended in your AI client or unattended on a schedule. The checkpoints you place hold either way.
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
Starting an agent on a timer takes a minute. Knowing it did the right thing, and what it cost, is the real work.
“Can I let it run while I sleep?”
Most tools make you choose: watch every move in a chat, or hand it to a script and hope.
“What did it do at 3am?”
A script on a timer leaves a log line on some box. Nobody can tell what ran, or why.
“What is this costing us?”
Runs nobody watches, with no cost per run, are a bill you discover at the end of the month.
Approving every step is one way to start, not a rule. Slide to see which steps wait for an approval, which run on their own, and what starts the run.
One way to start: you watch every step and give each one an explicit approval. Your AI tool is connected, and that is the whole setup.
drag, or use the arrow keys
Example: a bug sweep, run with you on Monday, then on its own on weekdays at 02:00, Berlin time.
In Claude Code you type "run the bug sweep". A task starts with its workflow attached. Nothing to install: your AI tool asks ConvOps for each step.
Payment webhook returns 500 first. OK?pick work: approve and continue?
example · approvals this run · 0 of 5
one approval left, before bugs are closed
Bug sweep · weekdays at 02:00
example data
You read what it picked, what it recalled and what it changed, and approve each step. Slow on purpose: one way to start, not a rule.
Payment webhook returns 500 first. OK?pick work: approve and continue?
example · approvals this run · 0 of 5
one approval left, before bugs are closed
Bug sweep · weekdays at 02:00
example data
The routine steps run straight through. You keep one approval, before bugs are closed. You could have kept all five, or none.
Payment webhook returns 500 first. OK?pick work: approve and continue?
example · approvals this run · 0 of 5
one approval left, before bugs are closed
Bug sweep · weekdays at 02:00
example data
Weekdays at 02:00 in Europe/Berlin. Each time it fires, a fresh task starts with the same workflow, the same rules and the same approval.
Payment webhook returns 500 first. OK?pick work: approve and continue?
example · approvals this run · 0 of 5
one approval left, before bugs are closed
Bug sweep · weekdays at 02:00
example data
The schedule fires. The run asks a pool for the next bug worth fixing and a runner you chose does the work.
Payment webhook returns 500 first. OK?pick work: approve and continue?
example · approvals this run · 0 of 5
one approval left, before bugs are closed
Bug sweep · weekdays at 02:00
example data
The close waited for you. The run left its log, its retries and its cost, on the task where you can read them.
Payment webhook returns 500 first. OK?pick work: approve and continue?
example · approvals this run · 0 of 5
one approval left, before bugs are closed
Bug sweep · weekdays at 02:00
example data
Payment webhook returns 500 first. OK?pick work: approve and continue?
example · approvals this run · 0 of 5
one approval left, before bugs are closed
Bug sweep · weekdays at 02:00
example data
A pool is a saved question about your work, answered live. Each run takes one task, or learns there is nothing to do.
Invoice exceptions waiting 3+ days
asking the pool
asking the pool
every run keeps its session, log, retries and cost
A pool keeps its own history: its size day by day, every pull and edit, and who pulls from it. A consumer that pulled and got nothing is called out.
Your AI tool asks for the step, does it, and moves on. No backend, no runner, nothing to deploy.
connect and go
Calendar rules, not cron strings: daily, weekdays, monthly on the 15th or a custom rule, each in its own time zone. Start a fresh task each time, or pick the same task back up where it stopped.
two modes
A saved question about your work, answered live. The run gets one task back, or nothing to do.
pulled, never pushed
Named profiles for where runs happen: your machines, ConvOps cloud, your own OpenCode server or your own runner.
swap without touching tasks
# weekdays at 2am server time, whatever the state
0 2 * * 1-5 agent --prompt "$(cat invoice-match.md)"RRULE (RFC 5545), not cron. Cadence kinds compile to a rule: once, every_minutes, every_hours, daily, weekdays, weekly, monthly or custom, read in the schedule's own IANA time zone and kept local across daylight saving. A preview shows the next five fires. Spawn creates a new child task each fire, carrying the whole workflow. Wake picks the same task back up at its current step.
{
"schedule_create": {
"title": "Supplier risk check {date}",
"rrule": "FREQ=WEEKLY;BYDAY=MO;BYHOUR=9",
"timezone": "Europe/Dublin"
},
"each fire": "spawns a new task"
}A pool is a typed, validated filter over tasks, resolved live on every call. pools_next returns exactly one task or an empty object, so a workflow can branch on emptiness.
{
"slug": "invoice-exceptions",
"filter": {
"op": "AND",
"nodes": [
{ "field": "tags", "contains": "invoice-exception" },
{ "field": "stale_days", "gte": 3 }
]
}
}Named execution profiles. ConvOps dispatches to an isolated run on Kubernetes or your OpenCode server, or your own runner claims the work over HTTPS and reports back. Precedence: step, then task, then workspace default.
{
"executors": [
{ "slug": "local", "is_system": true,
"description": "claude code on your machine" },
{ "slug": "opencode", "is_system": true },
{ "slug": "isolated-sonnet", "backend": "isolated",
"description": "one pod per run, your kubernetes" },
{ "slug": "gpu-box", "runner": "custom",
"description": "your infra" }
],
"precedence": "step > task > org default"
}The dispatch event is written in the same transaction as the task (outbox), redelivered idempotently, and orphaned runs are reaped. Each run records session id, execution log, retry count, status and cost in USD.
A ConvOps schedule starts the run, a pool can pick the work, and the workflow carries the process with whatever checkpoints you placed in it: an approval gate, a required metric, a required report, or none. The run executes on an executor you choose and records its transcript and cost. The same workflow runs attended in your AI client when you prefer to watch.
No. ConvOps runs attended by default: your AI client, connected over MCP, asks for the step, does the work and advances. A backend is only involved when a run must happen with nobody at the keyboard, and even then ConvOps cloud can host it.
A ConvOps schedule. You pick a cadence in plain words, such as weekdays at 08:00 or the 15th of every month, or write a custom RRULE, in the schedule's own time zone. A preview shows the next five fires in UTC and local time. Example: follow-the-sun triage, with one schedule each in Singapore, Dublin and Chicago.
It asks a ConvOps pool: a saved, typed query over tasks that the engine resolves live on every call. pools_next hands back exactly one task or nothing, and empty is an answer the workflow can branch on. Example: a pool of supplier invoices waiting more than three days, worked one at a time.
Wherever you point ConvOps: the AI clients on your machines, your own runner, which claims work over HTTPS and reports back, your own OpenCode server with your models, or ConvOps cloud. Execution is configured per workspace, so switching backends changes no task definition.
Yes, if the workflow has no approval gate. ConvOps checks the gates you placed and nothing else: a required metric or report can hold the close, and an approval gate holds a step until an explicit approval is given, recorded with the account that gave it. Add checkpoints where a person, a metric or a report should decide.
ConvOps writes the dispatch event in the same transaction as the task, so it cannot be lost. Delivery retries without duplicates, an unreachable backend is handled without breaking the task, and orphaned runs are cleaned up rather than left running.
ConvOps records each run's session, execution log, retry count and cost in dollars, and you can query all of it from your client. Example: a finance team compares the cost of its month-end reconciliation runs across model tiers before it picks one.