Control & safety
Autonomous, not unsupervised
Giving an AI team access to your code is a big decision. So the guardrails aren’t options you have to remember to turn on — they’re how Neeurn is built.
Content Writer needs your OK
Publish “How saved carts lift repeat orders” to WordPress
Go on — try the buttons.
The guardrails
Ten rules the system enforces, every time.
Never your default branch
Neeurn never pushes to a repository’s default branch, and checks the base branch on every pull request it opens. Changes always arrive as pull requests.
Agents can’t push
The AI works inside the container and can’t push or write outside the repository. The system pushes its branch after it finishes.
No credentials for the agents
Workers get model access and a short-lived, per-run permission token — never your GitHub, board or connector credentials.
Only the actions it’s allowed
Each run can call only the actions granted to it. Reviewers get read-only access. Nothing a document says can widen that.
Approval for risky actions
Destructive actions wait for your approval unless you allow them for a workflow; anything financial always needs approval.
Throwaway sandboxes
Every task runs in a fresh container that is destroyed afterwards, with capped CPU and memory, behind an egress firewall that blocks internal networks and cloud metadata.
Sealed secrets
Connection tokens are encrypted at rest with AES-256-GCM and decrypted only at the moment of use. Secrets are never written to logs.
Strict isolation
Every request is checked against your organization — including every id it mentions. One organization can never see or touch another’s data.
Untrusted content stays untrusted
Emails, web pages, tickets and documents are labeled as untrusted data for the AI; instructions hidden inside them can’t change what it may do.
Budgets that stop work
Each task has time, turn and cost limits, and each organization a monthly AI budget that stops new work before it’s crossed.
Your data
- Where it runs
- On our servers in Oracle Cloud. Code is only handled inside per-task containers and deleted with them; a cache of dependencies is kept per repository while your lockfile is unchanged.
- AI processing
- Tasks are processed by Anthropic’s Claude models through their API. Under Anthropic’s commercial terms, API data isn’t used to train their models.
- What we keep
- Your boards, tasks, run logs, reviews and the project memory your team builds — so the work and its history stay with you. Delete them, or your organization, any time.
- Your access
- Only members of your organization, with the roles and permissions you give them. Platform administrators can’t see inside your projects in the product.
Details are in our privacy policy. Found a security issue? Write to hello@neeurn.com.
Questions
Can a prompt hidden in an email or ticket make the AI do something else?
It can’t widen what the AI is allowed to do. Content from emails, web pages, tickets and documents is labeled as untrusted data, and each run can only call the actions in its own grant — decided by the workflow, not by what the AI reads or says.
Who can approve risky actions?
The workflow owner, from the task, the run page or the office. A workflow owner can allow specific destructive actions for one workflow; anything financial always needs approval, and no setting anywhere can change that.
Is our code used to train AI models?
No. Tasks are processed by Anthropic’s Claude models through their commercial API, under terms that don’t use API data for training. Your code is only handled inside a per-task container that is destroyed afterwards.
Can we run it on our own infrastructure?
Enterprise plans can run on dedicated servers or in your own cloud, with a DPA, a security review and an SLA. Write to us.
Keep exploring
The AI team →
Experts in every role, a Lead that routes, skills and working hours.
Engineering →
From a task on your board to a reviewed pull request.
Product & delivery →
Specs, stories, a clean backlog and sprints that plan themselves.
Content & marketing →
Positioning, SEO and articles drafted, edited and ready to publish.
Operations →
Briefs, inbox triage, reports and reminders — on a schedule.
Virtual office →
Watch work live; calls and meetings for your people.
Hire your AI team today.
Start free — no card, no interviews. Build your team in two minutes and watch it pick up its first task.