/pricing
Flat per org. Your models.
Planned prices are pre-launch hypotheses and may change. No annual plan.
Seats punish you for adding people, and they make no sense when the staff you're paying for isn't people.
Plans
Self-run
$199
Planned: hosted Management UI, your compute. Marshall, ephemeral agents, and the brain repo included. No limit on members or repos. 14-day trial. Compute and model usage are separate.
Get startedRepoStaff-hosted compute
$499
Planned price; may change. No member limit or per-seat charge. Hosted Management UI, RepoStaff's AWS. One Marshall host and 100 VM-hours included, then $0.60 per VM-hour up to your monthly spend limit ($100 by default, and yours to change).
Hosted compute waitlist detailsSelf-hoster rate
$49
Up to three members on their own hardware, still on the hosted Management UI. No customer AWS connection, email support, or trial.
Self-hoster rate detailsHosted management is not hosted compute. RepoStaff will host the Management UI on every plan. Worker compute is separate: your Linux/KVM hardware first (Coming soon), then your AWS account (Coming soon), then RepoStaff-hosted AWS (Coming soon; waitlisted, no date).
How the bill works.
Step 1 of 4
A flat subscription
Per org, not per seat, so adding another reviewer or approver costs nothing. It pays for the hosted Management UI, Marshall, ephemeral agents, and the workflow around the brain repo.
Step 2 of 4
Ongoing overhead
Persistent Marshall and shared infrastructure keep running between tasks, so those costs continue whether or not a ticket is moving.
Step 3 of 4
Included hosted usage
RepoStaff-hosted compute includes one Marshall host and 100 VM-hours a month. Self-run has no allowance, because the compute is yours.
Step 4 of 4
Then metered usage
Hosted VM-hours beyond the allowance are metered at a rate that includes a hosted-service margin. That margin is on compute. It isn't model-token resale.
A quiet month
An inactive period shouldn't generate task-usage charges for compute you don't consume. The subscription or included plan still continues, and so do persistent Marshall and shared infrastructure costs. Unused included hours don't refund or roll over.
What you pay someone else.
Your model provider
Model usage goes straight to Anthropic or OpenAI on your own API key, as a separate invoice. No subscription logins. RepoStaff doesn't resell tokens or mark them up. RepoStaff holds your keys in an encrypted vault and brokers requests by default, so they don't stay only on your machines, and bringing your own key isn't a no-retention promise.
Your compute
Self-run compute goes to whoever you already pay: your electric bill or your AWS bill.
Your AWS account
Connecting customer AWS starts persistent costs before any task runs, and they keep accruing without task activity: about $118 a month at us-east-1 list prices for network, proxy, and Marshall. That's an estimate, not a quote.
What the protections don't cover
RepoStaff's spend limits and budgets apply to RepoStaff's bill or stop runs based on estimates. They do not cap your model provider's invoice or your infrastructure invoices, and stopping runs doesn't stop always-on AWS charges.
You set the ceiling. RepoStaff stops at it.
Every hosted org starts with a $100 monthly spend limit on top of the plan. Lower it to $0, raise it to whatever your backlog needs, or add a daily limit to catch a runaway agent. When you reach a limit, runs stop and admins get an email. Raise it and they resume.
If RepoStaff is slow to stop, that compute is on RepoStaff. Your RepoStaff bill never goes past the limit you set.
Billing · Spend limits
Planned settings$100/ month · default
$100billed above the plan · runs stopped on day 23
Illustrative month- 50% alert emailed to admins ($50 of $100).
- 75% alert emailed to admins ($75 of $100).
- 90% alert emailed to admins ($90 of $100).
- Monthly limit reached. Runs stop. You're billed $100 above the plan, and no more.
The usage is invented to show how the limits behave: included hours run out on day 9, weekdays are steady after that, and an agent runs away on day 23. It isn't a forecast.
$100 is only the default.
Set $0 for included hours only, or thousands if your backlog needs it. The number is yours.
Warned before stopped.
Admins get an email at 50%, 75%, 90% and 100% of a limit, so it's never a surprise.
Never billed past it.
If RepoStaff is slow to stop, that compute is on RepoStaff, not on your invoice.
Spend limits are planned for hosted compute, which is waitlisted with no date. The settings above are a preview, not a live control. They cover your RepoStaff bill only. Model usage is billed by your provider on your own key, so set a limit there too; the Anthropic Console has one. On self-run, compute is billed by whoever runs it.
Limits, side by side.
Where the plans differ
- Agents run on
- Your Linux/KVM hardware or your AWS account
- Included compute
- None. The compute is yours
- After the allowance
- Not applicable. Self-run has no overage
- Monthly spend limit
- Not applicable. Self-run has no overage
- Daily limit
- Not applicable
- Workers
- Limited by your hardware or AWS account
- Fair use
- 1,500 runs a month. Two months above that starts a conversation, not a bill
- Org safety budget
- On by default, amount set by you. Reaching it stops new and active runs
- Free trial
- 14 days, one per org, card required, not charged before day 15
- Support
- Email, no response-time guarantee
The same on both plans
- Management UI. Hosted by RepoStaff
- Marshall and the brain repo. One org, one Marshall, included
- Members and repos. No limit on either
- Per-ticket budget. Set by you. Reaching it stops that run and hands the ticket to a person
- Model keys. Yours, Anthropic or OpenAI
- Billing. Monthly. No annual plan
On hosted compute, the most you can owe RepoStaff in a month is $499 plus your monthly spend limit. Your model provider's invoice sits outside that figure.
What the whole bill might look like.
Model spend is usually the largest line, so RepoStaff's price alone is the wrong number to compare. Your total is three lines: the RepoStaff fee, your model provider's invoice, and your compute. These are modeled from list prices, not measured from customer runs.
Modeled, not measured
Modeled monthly totals in US dollars, including the RepoStaff price. Estimates, not quotes. Bars share one scale.
Light40 runs a month
Self-run, your hardware$442
Self-run, your AWS$568
RepoStaff-hosted compute$742
Typical120 runs a month
Self-run, your hardware$928
Self-run, your AWS$1,079
RepoStaff-hosted compute$1,228
Heavy400 runs a month
Self-run, your hardware$2,589
Self-run, your AWS$2,863
RepoStaff-hosted compute$3,069
| Usage per month | Model spend | Self-run, your hardware | Self-run, your AWS | RepoStaff-hosted compute |
|---|---|---|---|---|
| Light: 40 runs | $243 | $442 | $568 | $742 |
| Typical: 120 runs | $729 | $928 | $1,079 | $1,228 |
| Heavy: 400 runs | $2,390 | $2,589 | $2,863 | $3,069 |
Model assumptions: list prices from 2026-11-08, workers on Claude Sonnet 5.5 and Marshall on Claude Opus 5.5. Token volumes are estimates. Your own hardware adds electricity. The heavy hosted figure includes $180 of overage, so it needs a monthly spend limit of at least $180; the default is $100. Seat-based tools usually include some model usage, so compare totals and not list prices.
Two things RepoStaff won't sell you.
A free hosted trial
A free VM with a browser attracts exactly who you'd expect, and hosted compute costs real money from minute one. It needs a card on file, and every org starts with a spend limit.
An enterprise plan
There isn't one: no SOC 2, no procurement portal, no "contact sales" hiding the price. If you need those, buy from someone who has them.
Ready when you are.
If the Management UI says it isn't worth it, cancel. The knowledge committed to your brain repo stays with you. Application history, session archives, credentials, and live services need separate export or migration.
Read the FAQ for the boundaries on data, models, and compute.