/faq
Questions worth asking.
Pre-launch · Coming soon. These answers describe planned behavior, not an operational service.
The basics
How do I get started?
Coming soon; onboarding is not open. Subscribe for the launch email, or email support@repostaff.com with questions. Nothing touches your code before a signed agreement.
What are the three pillars?
Ephemeral agents will implement and validate scoped work. One renameable Marshall per organization will plan, triage, delegate, and check. The Management UI will oversee work, connections, approvals, spending, and outcomes. The brain repo you own supports all three.
Will this replace my engineers?
No, and RepoStaff isn't sold that way. It's for a team you already have, to add capacity for backlog work the team wants done and can't currently staff.
Is Marshall as good as a staff engineer?
That's for your team to judge. Marshall has that job, and the Management UI shows you how it's going.
Should I just buy Devin?
Maybe. As far as I could tell on 2026-11-08, running it in your own VPC takes their Enterprise plan, and what it learns lives inside their product. If neither bothers you, go buy it.
Why should I trust one person with this?
You shouldn't have to. Self-run keeps the agents on infrastructure you pick, and the brain is already yours. If RepoStaff disappears, you still have the repo. Weigh the rest for yourself.
Work and approvals
How do approvals work?
Cautious is the default; Balanced and Hands-off are alternatives. Human, agent, and CI gates will govern the fixed flow from triggered to merged, and you can set your own. Policy lives in the brain repo, and Management UI changes will become commits. Workers will read pinned brain snapshots and will not be able to rewrite policy or broaden their own permissions.
Does a chat request start work?
The plan: an authorized chat request will create proposed work, and Marshall can draft the ticket. A proposal doesn't authorize execution. That still needs the configured authorization and gates.
Can someone who doesn't write code start work?
They can ask Marshall for something in plain language, and it drafts the ticket. A developer decides when the work starts, unless your team has set it up differently.
What can a worker do?
Run your tests, bring up your app in Docker, use it in a browser, and attach evidence to the pull request: a test summary, plus screenshots or a recording when the UI changed.
What can a worker reach?
Short-lived scoped credentials, restricted network access by default, a read-only copy of the brain, and no production credentials.
Does evidence guarantee the change is right?
No. Tests and relevant screenshots or recordings accompany the pull request to support review. They do not guarantee correctness.
Can the agents deploy to production?
No. The workflow ends at merge, not deployment. Agents get no production credentials, and nothing here deploys for you.
Can I stop a run?
The plan is to let you cancel a run from the Management UI, retry it, send it back with a comment, or abandon it. Budgets and per-ticket limits will stop runs too.
The brain and Marshall
What's in the brain repo?
org/, decisions/, memories/, skills/, and policy/. Plain files in a git repo in your org.
What do I keep if I leave?
The sourced knowledge committed to your brain repo remains yours. Cloning it does not export application history, session archives, credentials, or live services. Those need separate export or migration, so take what you need before you go.
Can I edit what Marshall remembers?
The plan is plain files you can edit and push.
How do you stop a bad memory from doing damage?
The plan is that every memory will record where it came from. Agents get a read-only copy and treat memories as information, never as commands. Untrusted input goes to quarantine by default, which each org can switch off, and Marshall can't loosen its own permissions without a human. No defense against deliberately poisoned input is perfect, so review what matters.
Who approves changes to the brain?
You set that per org. The default lets memories through without approval and asks for it on skills/ and policy/. Loosening that is an explicit, audited choice.
How does Marshall learn my org?
Once onboarding opens, an admin will start it. Marshall will read your code, docs, CI setup, recent merged pull requests, and tickets, will confirm what it worked out, and will interview you on what it couldn't. Your answers will become commits in the brain, with their source.
Can I rename Marshall?
Yes. One Marshall per org.
Connections, models and compute
What does it connect to?
Coming soon: GitHub for code, CI, Issues and the brain repo, with Discord first. Linear and Slack will follow, then Trello.
Can you add a provider I need?
No provider or date is promised, but ask: support@repostaff.com
Do you support GitLab, Bitbucket, or Azure DevOps?
No. GitHub is the only code host and CI.
Do you support Jira or Microsoft Teams?
No. Neither is supported yet.
Where does it run?
Coming soon: RepoStaff will host the Management UI. Planned agent compute is separate: Linux/KVM hardware first, then customer AWS, then gated RepoStaff-hosted AWS. All are Coming soon. Hosted compute has no promised date.
Can I self-host the whole service?
No. Customer-run agent compute is a separate option. Full management self-hosting has no date, easy setup, or support promised.
Which models does it use?
Anthropic and OpenAI, through your own API keys.
Can I use my Claude or ChatGPT subscription?
No. Subscription logins aren't a safe basis for a product like this. API keys only, and RepoStaff doesn't resell model tokens.
Data, security and source
Where does my code go?
Running agents on your hardware doesn't keep your code entirely on your network. Selected context goes to your model provider, and source may pass through the hosted control plane. Each worker is one fresh VM for one ticket.
Will my data stay in the US?
That isn't guaranteed. RepoStaff serves US customers and primary hosting is in the US, but neither is a promise that all data stays in the US.
Where do my model keys go?
To RepoStaff. By default you submit each key to an encrypted vault, and agents reach your provider through a broker with a short-lived token for that job. Delivering a key directly into a job is off by default. Your provider still receives task and code context, so bringing your own key isn't a no-retention promise.
What runs on RepoStaff's servers if I self-run?
The control plane: the Management UI and the integrations. Chat turns Marshall accepts and brain writes pass through it, and source and prompts may too.
Will Marshall read all our channels?
No. It works in DMs and in threads it's mentioned into. Chat platforms still send RepoStaff other channel events, and RepoStaff discards those without storing them or passing them to Marshall.
How long do you keep our conversations?
Ask before you connect a repo: support@repostaff.com
Do you have SOC 2?
No. There's no penetration test or completed audit either. If you need those today, RepoStaff isn't ready for you.
Is RepoStaff open source?
No. Paying customers receive source access and an issues and PR path. The brain is your own repo, so that's yours regardless.
Cost and contact
What does it cost?
Planned prices are pre-launch hypotheses and may change. Flat per org, not per seat. $199 a month for self-run and $499 a month for RepoStaff-hosted compute, plus your own model usage. See pricing.
Why per org, not per seat?
Because the people who review and approve the work are the point. A seat price would charge you for every reviewer you add. Per org, adding reviewers and approvers costs nothing.
What will my total bill be?
Three lines: the RepoStaff fee, your model provider's invoice on your own keys, and your compute. Model spend is usually the largest. See the modeled totals, which are estimates from list prices, not measured runs.
Are these prices final?
No. They're pre-launch hypotheses, not yet tested with buyers, and may change before launch.
Can I cap what RepoStaff bills me?
Yes. On hosted compute, every org starts with a $100 monthly spend limit on overage above the plan. Set any amount from $0, which means included hours only, add a daily limit to catch a runaway agent, or turn it off. Admins get an email at 50%, 75%, 90% and 100%, and reaching a limit stops new and active runs until you raise it. If RepoStaff is slow to stop, that compute is on RepoStaff. Spend limits are planned with hosted compute and don't cover your model provider's invoice. See how limits work.
What do I pay in a quiet month?
An inactive period shouldn't generate task-usage charges for compute you don't consume. The subscription or included plan continues, and so do persistent Marshall and shared infrastructure costs, including any customer AWS costs.
Is there a cheaper option for self-hosters?
A $49 rate is coming later for orgs of up to three members on their own hardware, with community help and no email support. It opens only if the support load allows. Read the self-hoster conditions.
How do I reach you?
support@repostaff.com. Support is by email, with no response-time guarantee.