unattended AI agent runs: It runs overnight. You read the result.

Unattended runs are ConvOps workflows that start on a schedule, without anyone typing. Every run is on the record. People keep the checkpoints they placed and the stop button.

schedules · one pod per run · tokens and cost per run · gates still hold

000306091215182101:0002:0003:0004:30
22:30sun · Europe/Berlinnight window
schedulesEurope/Berlin
  • 02:00Bug sweepweekdays 02:00not today
  • 03:00Supplier invoice matchevery day 03:00in 4h 30m
  • 04:30Weekly SEO auditmondays 04:30not today
next: Content draft in 2h 30m
convops · runs tonightexample
runs0
live0
tokens0
spend$0.00
done0/0
taskexecutorstatustokenscosttime
Nightly content draftclaude-sonnet running0$0.000m 00s
Bug sweepclaude-opus running0$0.000m 00s
Supplier invoice matchclaude-haiku running0$0.000m 00s
Weekly SEO auditopencode-kimi running0$0.000m 00s
spend by model
opus $0.00 sonnet $0.00 kimi $0.00 haiku $0.00
in short

How do you run AI agents unattended?

Give the workflow a schedule. Each run starts on its own, works in its own pod, and the work lands on a review branch, or any branch you choose.

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

  • Four things can start a run without typing: a schedule, a run dispatching a follow-up task, an automatic resume after a failed run, and your own app through the SDK.
  • A failed run resumes on its own on the same volume, and the record links both runs. A run that reaches its time limit (four hours by default) keeps its files and is retried by hand.
  • Work is pushed to a task branch for review, or to a branch you choose. When a push cannot land, the work goes to a rescue branch, so nothing the run did is lost.
  • Approval gates hold whether a person or a schedule started the run.

best for

  • Overnight or weekly agent work you want to read in the morning.
  • Teams that need a cost and a record for every unattended run.

not for

  • Shipping with nobody accountable. Where the workflow has an approval gate, the run holds at that step until an explicit approval is given, recorded with the account that gave it.

updated

the problem

The work stops when you log off.

AI that only runs while someone types is a tool, not a team. And what runs unwatched is rarely on any record.

“Why does the AI only work when someone types?”

Every run needs a person at a keyboard and a chat left open. When they log off, the work stops.

“What ran last night?”

A script fired on someone's machine. Whether it finished, failed or never started, nobody can say.

“What did it cost?”

Tokens disappear into one monthly bill. No run, no model and no task carries its own number.

one night

From 01:00 to your morning coffee.

Example: one nightly bug sweep, start to finish. Nobody typed a word until the review.

01the schedule

01:00. The schedule fires.

The bug sweep has a rule: every weekday at 01:00 in Europe/Berlin. Nobody is awake. The schedule creates the task and starts the run.

000306091215182101:00
22:30sun · Europe/Berlinnight window
schedulesEurope/Berlin
  • 01:00Bug sweepweekdays 01:00 · Europe/Berlinnot today
next: Bug sweep in 2h 30m
cadence · exampleweekdays at 01:00 · Europe/Berlin · next five runs shown before saving
namespaceconvops-runs
podrun-8c1epending
main
agentclaude code
step done
sidecar
mcptool sidecar
task volumekept across stepskept
run secretthis run only
run-8c1esucceededstep donerun recorded
  • non-root
  • no SA token
  • caps dropped
  • restricted
runBug sweep 2026-10-02scheduled
queuedpod startingworkingresumedpushing
calling
pushed
tokens0k
cost$0.00
duration0h 0m
a failed run · resumed on its own

The run ends as a failure. The task volume, and every file on it, stays. Three failures in a row and the task waits for you.

run_4f2a · resumeexample
main
Review branchPushed for review on convops/task/8c1e.
convops · last nightexample
runs0
live0
tokens0
spend$0.00
done0/0
taskexecutorstatustokenscosttime
Bug sweep · node lostclaude-opus running0$0.000m 00s
Bug sweep · resumedclaude-opus running0$0.000m 00s
Nightly content draftclaude-sonnet running0$0.000m 00s
spend by model
opus $0.00 sonnet $0.00
example · your AI client · reviewwaiting for approval

Bug sweep 2026-10-02

step 5 of 6 · review

convops/task/8c1e

runs
2 · resumed once
changes
7 files · tests pass
cost
$5.18 · opus
Reply "approved" Ask for changes
02the run

The run gets its own pod.

One pod for this run: the agent, its tools beside it, a volume for the task and a secret for this run only. It works the workflow step by step.

000306091215182101:00
22:30sun · Europe/Berlinnight window
schedulesEurope/Berlin
  • 01:00Bug sweepweekdays 01:00 · Europe/Berlinnot today
next: Bug sweep in 2h 30m
cadence · exampleweekdays at 01:00 · Europe/Berlin · next five runs shown before saving
namespaceconvops-runs
podrun-8c1epending
main
agentclaude code
step done
sidecar
mcptool sidecar
task volumekept across stepskept
run secretthis run only
run-8c1esucceededstep donerun recorded
  • non-root
  • no SA token
  • caps dropped
  • restricted
runBug sweep 2026-10-02scheduled
queuedpod startingworkingresumedpushing
calling
pushed
tokens0k
cost$0.00
duration0h 0m
a failed run · resumed on its own

The run ends as a failure. The task volume, and every file on it, stays. Three failures in a row and the task waits for you.

run_4f2a · resumeexample
main
Review branchPushed for review on convops/task/8c1e.
convops · last nightexample
runs0
live0
tokens0
spend$0.00
done0/0
taskexecutorstatustokenscosttime
Bug sweep · node lostclaude-opus running0$0.000m 00s
Bug sweep · resumedclaude-opus running0$0.000m 00s
Nightly content draftclaude-sonnet running0$0.000m 00s
spend by model
opus $0.00 sonnet $0.00
example · your AI client · reviewwaiting for approval

Bug sweep 2026-10-02

step 5 of 6 · review

convops/task/8c1e

runs
2 · resumed once
changes
7 files · tests pass
cost
$5.18 · opus
Reply "approved" Ask for changes
03a failure

03:10. The node goes away.

Halfway through the sweep, the machine under the run is lost. The run ends as a failure. Nothing is thrown away: the volume holds every file it touched.

000306091215182101:00
22:30sun · Europe/Berlinnight window
schedulesEurope/Berlin
  • 01:00Bug sweepweekdays 01:00 · Europe/Berlinnot today
next: Bug sweep in 2h 30m
cadence · exampleweekdays at 01:00 · Europe/Berlin · next five runs shown before saving
namespaceconvops-runs
podrun-8c1epending
main
agentclaude code
step done
sidecar
mcptool sidecar
task volumekept across stepskept
run secretthis run only
run-8c1esucceededstep donerun recorded
  • non-root
  • no SA token
  • caps dropped
  • restricted
runBug sweep 2026-10-02scheduled
queuedpod startingworkingresumedpushing
calling
pushed
tokens0k
cost$0.00
duration0h 0m
a failed run · resumed on its own

The run ends as a failure. The task volume, and every file on it, stays. Three failures in a row and the task waits for you.

run_4f2a · resumeexample
main
Review branchPushed for review on convops/task/8c1e.
convops · last nightexample
runs0
live0
tokens0
spend$0.00
done0/0
taskexecutorstatustokenscosttime
Bug sweep · node lostclaude-opus running0$0.000m 00s
Bug sweep · resumedclaude-opus running0$0.000m 00s
Nightly content draftclaude-sonnet running0$0.000m 00s
spend by model
opus $0.00 sonnet $0.00
example · your AI client · reviewwaiting for approval

Bug sweep 2026-10-02

step 5 of 6 · review

convops/task/8c1e

runs
2 · resumed once
changes
7 files · tests pass
cost
$5.18 · opus
Reply "approved" Ask for changes
04auto-resume

It resumes where it stopped.

ConvOps starts a new run on the same volume, on its own, and it carries on. The record links it to the run it continued, so the night reads as one story.

000306091215182101:00
22:30sun · Europe/Berlinnight window
schedulesEurope/Berlin
  • 01:00Bug sweepweekdays 01:00 · Europe/Berlinnot today
next: Bug sweep in 2h 30m
cadence · exampleweekdays at 01:00 · Europe/Berlin · next five runs shown before saving
namespaceconvops-runs
podrun-8c1epending
main
agentclaude code
step done
sidecar
mcptool sidecar
task volumekept across stepskept
run secretthis run only
run-8c1esucceededstep donerun recorded
  • non-root
  • no SA token
  • caps dropped
  • restricted
runBug sweep 2026-10-02scheduled
queuedpod startingworkingresumedpushing
calling
pushed
tokens0k
cost$0.00
duration0h 0m
a failed run · resumed on its own

The run ends as a failure. The task volume, and every file on it, stays. Three failures in a row and the task waits for you.

run_4f2a · resumeexample
main
Review branchPushed for review on convops/task/8c1e.
convops · last nightexample
runs0
live0
tokens0
spend$0.00
done0/0
taskexecutorstatustokenscosttime
Bug sweep · node lostclaude-opus running0$0.000m 00s
Bug sweep · resumedclaude-opus running0$0.000m 00s
Nightly content draftclaude-sonnet running0$0.000m 00s
spend by model
opus $0.00 sonnet $0.00
example · your AI client · reviewwaiting for approval

Bug sweep 2026-10-02

step 5 of 6 · review

convops/task/8c1e

runs
2 · resumed once
changes
7 files · tests pass
cost
$5.18 · opus
Reply "approved" Ask for changes
05the run

It pushes the task branch.

The fixes go to a branch named for the task, ready for review; the environment could name any branch you choose instead. Had the push been refused, the work would sit on a rescue branch.

000306091215182101:00
22:30sun · Europe/Berlinnight window
schedulesEurope/Berlin
  • 01:00Bug sweepweekdays 01:00 · Europe/Berlinnot today
next: Bug sweep in 2h 30m
cadence · exampleweekdays at 01:00 · Europe/Berlin · next five runs shown before saving
namespaceconvops-runs
podrun-8c1epending
main
agentclaude code
step done
sidecar
mcptool sidecar
task volumekept across stepskept
run secretthis run only
run-8c1esucceededstep donerun recorded
  • non-root
  • no SA token
  • caps dropped
  • restricted
runBug sweep 2026-10-02scheduled
queuedpod startingworkingresumedpushing
calling
pushed
tokens0k
cost$0.00
duration0h 0m
a failed run · resumed on its own

The run ends as a failure. The task volume, and every file on it, stays. Three failures in a row and the task waits for you.

run_4f2a · resumeexample
main
Review branchPushed for review on convops/task/8c1e.
convops · last nightexample
runs0
live0
tokens0
spend$0.00
done0/0
taskexecutorstatustokenscosttime
Bug sweep · node lostclaude-opus running0$0.000m 00s
Bug sweep · resumedclaude-opus running0$0.000m 00s
Nightly content draftclaude-sonnet running0$0.000m 00s
spend by model
opus $0.00 sonnet $0.00
example · your AI client · reviewwaiting for approval

Bug sweep 2026-10-02

step 5 of 6 · review

convops/task/8c1e

runs
2 · resumed once
changes
7 files · tests pass
cost
$5.18 · opus
Reply "approved" Ask for changes
06you approve

08:00. You read it and decide.

Two runs, the tokens, the cost per model, the duration. The workflow waits at the review gate until an explicit approval is given. You read the branch and give it, recorded with your account.

000306091215182101:00
22:30sun · Europe/Berlinnight window
schedulesEurope/Berlin
  • 01:00Bug sweepweekdays 01:00 · Europe/Berlinnot today
next: Bug sweep in 2h 30m
cadence · exampleweekdays at 01:00 · Europe/Berlin · next five runs shown before saving
namespaceconvops-runs
podrun-8c1epending
main
agentclaude code
step done
sidecar
mcptool sidecar
task volumekept across stepskept
run secretthis run only
run-8c1esucceededstep donerun recorded
  • non-root
  • no SA token
  • caps dropped
  • restricted
runBug sweep 2026-10-02scheduled
queuedpod startingworkingresumedpushing
calling
pushed
tokens0k
cost$0.00
duration0h 0m
a failed run · resumed on its own

The run ends as a failure. The task volume, and every file on it, stays. Three failures in a row and the task waits for you.

run_4f2a · resumeexample
main
Review branchPushed for review on convops/task/8c1e.
convops · last nightexample
runs0
live0
tokens0
spend$0.00
done0/0
taskexecutorstatustokenscosttime
Bug sweep · node lostclaude-opus running0$0.000m 00s
Bug sweep · resumedclaude-opus running0$0.000m 00s
Nightly content draftclaude-sonnet running0$0.000m 00s
spend by model
opus $0.00 sonnet $0.00
example · your AI client · reviewwaiting for approval

Bug sweep 2026-10-02

step 5 of 6 · review

convops/task/8c1e

runs
2 · resumed once
changes
7 files · tests pass
cost
$5.18 · opus
Reply "approved" Ask for changes
000306091215182101:00
22:30sun · Europe/Berlinnight window
schedulesEurope/Berlin
  • 01:00Bug sweepweekdays 01:00 · Europe/Berlinnot today
next: Bug sweep in 2h 30m
cadence · exampleweekdays at 01:00 · Europe/Berlin · next five runs shown before saving
namespaceconvops-runs
podrun-8c1epending
main
agentclaude code
step done
sidecar
mcptool sidecar
task volumekept across stepskept
run secretthis run only
run-8c1esucceededstep donerun recorded
  • non-root
  • no SA token
  • caps dropped
  • restricted
runBug sweep 2026-10-02scheduled
queuedpod startingworkingresumedpushing
calling
pushed
tokens0k
cost$0.00
duration0h 0m
a failed run · resumed on its own

The run ends as a failure. The task volume, and every file on it, stays. Three failures in a row and the task waits for you.

run_4f2a · resumeexample
main
Review branchPushed for review on convops/task/8c1e.
convops · last nightexample
runs0
live0
tokens0
spend$0.00
done0/0
taskexecutorstatustokenscosttime
Bug sweep · node lostclaude-opus running0$0.000m 00s
Bug sweep · resumedclaude-opus running0$0.000m 00s
Nightly content draftclaude-sonnet running0$0.000m 00s
spend by model
opus $0.00 sonnet $0.00
example · your AI client · reviewwaiting for approval

Bug sweep 2026-10-02

step 5 of 6 · review

convops/task/8c1e

runs
2 · resumed once
changes
7 files · tests pass
cost
$5.18 · opus
Reply "approved" Ask for changes
what starts a run

Four ways work starts on its own.

None of them needs a person at a keyboard. All of them land on the same record.

A schedule

A recurrence rule on the task, such as every weekday at 01:00 in Europe/Berlin, in its own time zone. You can also fire it by hand.

rule-based

A run's follow-up

A run can dispatch the next task it finds, within a per-run limit, so one night's work can lead to the next.

bounded

A resume

A run that fails starts again on its own, on the same volume, and carries on from where it stopped.

same volume

Your app

Your own product creates scheduled work through the SDK, when your customers or your data call for it.

through the SDK

schedules

Rules you can read out loud.

Daily, weekdays, weekly, monthly or a custom rule, each schedule in its own time zone. See the next five runs before you save, and fire it by hand.

000306091215182101:0002:0003:3005:00
22:30sun · Europe/Berlinnight window
schedulesEurope/Berlin
  • 02:00Bug sweepweekdays 02:00not today
  • 03:30Sanctions list checkevery day 03:30in 5h 00m
  • 05:00Weekly SEO auditmondays 05:00not today
next: Content draft in 2h 30m
history and cost

Every run, on the record.

The Executions overview shows run status, runs per day, runs by executor and by backend, and spend. Every run keeps its tokens, cost per model and duration.

convops · executionsexample
runs0
live0
tokens0
spend$0.00
done0/0
taskexecutorstatustokenscosttime
Nightly content draftclaude-sonnet running0$0.000m 00s
Bug sweepclaude-opus running64k$1.295m 36s
Weekly SEO auditopencode-kimi running0$0.000m 00s
Sanctions list checkclaude-sonnet running0$0.000m 00s
Month-end accrual draftclaude-sonnet running0$0.000m 00s
Weekly KPI reportclaude-haiku running0$0.000m 00s
spend by executor
claude-opus $0.00 claude-sonnet $0.00 opencode-kimi $0.00 claude-haiku $0.00
overview
run status, runs per day, by executor, by backend
spend
last 7 days, and its trend
tokens · cost
per model, reported or computed, and says which
how it ended
model, infrastructure, time limit, stopped
how runs end

Four endings. Nothing lost.

Finished, stopped by a person, resumed after a failure, or failed with the work kept. Each one is recorded.

runNightly content draftscheduled
queuedpod startingworkingfinishing
calling
pushed
tokens0k
cost$0.00
duration0m 0s
runSanctions list checkscheduled
queuedpod startingworkingfinishing
calling
stopped
tokens0k
cost$0.00
duration0m 0s
runBug sweepscheduled
queuedpod startingworkingresumedfinishing
calling
pushed
tokens0k
cost$0.00
duration0h 0m
runWeekly KPI reportscheduled
queuedpod startingworkingfinishing
calling
rescue branch
tokens0k
cost$0.00
duration0m 0s

where the work lands

main
Task branchPushed for review on its own task branch.
retry

Retry a failed run, your way.

A failed, stopped or timed-out run can be retried by hand. Resume carries on with the working copy it left; Start fresh begins clean.

runWeekly KPI report · run 1scheduled
queuedpod startingworkingfinishing
calling
work kept
tokens0k
duration0m 0s

Failed. The working copy stays.

runWeekly KPI report · run 2scheduled
queuedpod startingworkingfinishing
calling
pushed
triggermanual retry
duration0m 0s

Retried with Resume. Finished.

  • ResumeKeeps the working copy as the failed run left it, and carries on.
  • Start freshRebuilds the working copy from the last pushed state, then starts clean.
  • From the task, the Executions list or the API.
  • A stalled or blocked task goes back to in progress.
  • Refused while a run is live, or when the same failure would repeat. The reason is named.
agent git access

The agent pushes its own work.

Give an environment full git access and the agent pushes as it goes, fast-forward only. Work lands on a review branch, or any branch you choose.

run log · gitexample
  • By default, the agent only commits.ConvOps pushes the work when the run ends.
  • With full git access, it pushes its own work.Fast-forward only: no force-push, no other branches, no deleted branches.
  • The destination comes from the environment.A review branch, or any branch you choose. Never from the prompt.
  • A refused push is explained.The reason and the fix are on the run, and the task waits for you instead of failing the same way again.
How git access works
follow-ups

One run can start the next.

A run can dispatch follow-up tasks it finds, in its own workspace, when its environment allows it. Each run has a limit, so one night cannot run away.

one run

Three invoice exceptions found. Dispatch one task for each.

Match invoice 4471dispatched · 1 of 10
Chase missing PO 8812dispatched · 2 of 10
Review Kessler duplicatedispatched · 3 of 10
Match invoice 4471follow-up task
Chase missing PO 8812follow-up task
Review Kessler duplicatefollow-up task
limit
10 per run by default, set per environment from 0 to 100
past the limit
refused, and the run is told why
where
tasks of the run's own workspace, never its own task
on the record
each follow-up names the run that dispatched it
the gate

Checkpoints hold while you sleep.

A schedule starts the work, it does not approve it. A checkpoint you placed still holds: an approval, a required metric or a report. Any run can be stopped.

schedule01:00 Berlin
workown pod
pushtask branch
reviewapproval
merge
example · your AI client · reviewwaiting for approval

Bug sweep 2026-10-02

step 5 of 6 · review

convops/task/8c1e

runs
2 · resumed once
changes
7 files · tests pass
cost
$5.18 · opus
Reply "approved" Ask for changes

example · the gate holds the step until an explicit approval is given, recorded with the account that gave it

runExample: reconcile supplier invoicesscheduled
queuedpod startingworkingfinishing
calling
stopped
tokens0k
cost$0.00
duration0m 0s

checkpoints you choose

Add checkpoints where a person, a metric or a report should decide. Leave them out where the work can run by itself.

Attended or on its own
the difference

A laptop left open vs a record.

  • A script on a timer, on one machine someone has to keep awake.
  • The run shares a laptop with everything else on it.
  • A job that crashes halfway loses its work.
  • Output lands wherever the script wrote it.
  • Nobody knows what it cost until the invoice.
  • It ships straight through. Nobody decided.
  • A schedule on the task, in its own time zone, with its cadence described back in plain words.
  • Each run gets its own pod, its own volume and its own secrets.
  • A failed run resumes on the same volume and carries on where it stopped.
  • Work is pushed to a task branch or the branch you choose, or kept on a rescue branch when the push cannot land.
  • Tokens, per-model cost and duration are recorded for every run.
  • The checkpoints you placed still hold: an approval, a required metric or a report.

A schedule is a recurrence rule

An RFC 5545 rule, or a cadence in plain words, on a scheduled task. Hours are read in the schedule's own IANA time zone and stay local across daylight saving. Each fire creates a child task that inherits the title, project and executor, then runs it.

payload
{
  "tool": "schedule_create",
  "arguments": {
    "title": "Bug sweep {date}",
    "vertical": "development",
    "project": "checkout",
    "rrule": "FREQ=WEEKLY;BYDAY=MO,TU,WE,TH,FR;BYHOUR=1;BYMINUTE=0",
    "timezone": "Europe/Berlin",
    "executor": "claude-opus",
    "description": "Pick open bugs, fix, test, push the task branch."
  }
}
schedule_create · example

Every run is a record

Each run keeps its trigger, outcome, error class and stats: tokens in total and per model, turns, duration and cost, with where the cost came from.

payload
{
  "id": "run_4f2a",
  "executor_name": "claude-opus",
  "agent": "claude-code",
  "trigger": "auto_resume",
  "resume_of": "run_9b71",
  "status": "ended",
  "outcome": "succeeded",
  "stats": {
    "tokens": 71000,
    "tokens_by_model": {
      "opus": { "tokens": 64200, "cost_usd": 0.79 },
      "haiku": { "tokens": 6800, "cost_usd": 0.03 }
    },
    "turns": 38,
    "duration_ms": 1330000,
    "cost_usd": 0.82,
    "cost_source": "computed"
  }
}
task execution record · example

The run before it

The first run of the night ended when its node was lost. Its outcome and error class show it, and the resume above points back to it.

payload
{
  "id": "run_9b71",
  "trigger": "dispatch",
  "status": "ended",
  "outcome": "infra_failure",
  "error_class": "node_lost",
  "stats": {
    "tokens": 341000,
    "tokens_by_model": { "opus": { "tokens": 341000, "cost_usd": 4.36 } },
    "turns": 214,
    "duration_ms": 7740000,
    "cost_usd": 4.36,
    "cost_source": "reported"
  }
}
task execution record · example

Time limit and follow-ups

The time limit is set per executor, four hours by default. A run that reaches it ends as time_limit and is retried by hand. A run may dispatch follow-up tasks, within a per-run limit, and any run can be stopped.

payload
{
  "name": "claude-opus",
  "backend": "isolated",
  "isolated": {
    "agent": "claude-code",
    "model": "opus",
    "time_limit_s": 14400
  }
}
executor · example
questions

Unattended run questions, answered.

What is an unattended run in ConvOps?

An unattended run is a ConvOps workflow run that starts with nobody typing: on a schedule, as a follow-up from another run, as an automatic resume or from your own app. Each run gets its own isolated pod, records its tokens and cost, and stops at the checkpoints you placed. Example: a supplier invoice match runs every weekday at 03:00 in Europe/Berlin.

What can start a run without anyone typing?

Four things in ConvOps. A schedule, such as every weekday at 01:00 in Europe/Berlin or the 15th of every month. A run dispatching a follow-up task, within a per-run limit. An automatic resume after a failed run. And your own app, creating scheduled work through the SDK. You can also trigger any schedule by hand, which does not move its next fire.

What happens when a run fails or hits its time limit?

In ConvOps a run that fails, say the model errors or the node under it is lost, resumes on its own on the same volume, and the record links the two runs. After three failed runs in a row the task stops and waits for a person. A run that reaches its time limit, four hours by default, stops and keeps its files: retry it with Resume and it carries on.

Where does the work end up?

ConvOps pushes to a task branch for review by default, or to any branch you choose in the environment, where the work lands only when the run succeeded with a clean tree. When a push cannot land, the work goes to a rescue branch, so nothing the run did is lost. Reports, drafts and data files land the same way as code.

Can an unattended run finish without a person?

Only if your ConvOps workflow lets it. The checkpoints in the workflow hold whether a person or a schedule started the run: an approval gate holds the 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. Leave them out where the work can run by itself.

How do we know what a run cost?

ConvOps records every run with its tokens, a per-model split, cost, turns, duration and, when it fails, an error class. Cost is either reported by the agent or computed from tokens, and the record says which. Example: a compliance team compares what each nightly sanctions check costs on two model tiers.

Can we stop a run that is going wrong?

Yes. Any running ConvOps run can be stopped from the runs view or through the API. The record keeps it as stopped, with what it used up to that point, and a person can retry it later with Resume or Start fresh.

Do runs need our laptops or our servers?

No laptop. ConvOps runs execute in isolated pods, one per run, on our cloud or on your own Kubernetes cluster. Enterprise customers can self-host from one Helm chart delivered with Enterprise, so nothing has to leave the cluster. Example: a hospital group keeps its nightly claims check inside its own cluster.