A private machine per project
Every project runs in its own isolated virtual machine — not a shared container or browser tab. Your code, dependencies, and data stay walled off from every other project.
Trust & security
Fast doesn’t have to mean reckless.
Gate 15 keeps the AI out in the open. It shows its plan before making changes, keeps your keys encrypted, isolates every project, and only publishes when you say so.
1 VM
per project — fully isolated
AES-256
encryption for secrets at rest
TLS 1.3
in transit, everywhere
0
secret values shown back to the model
How we protect your work
Security built into the workflow.
These aren’t bolted-on settings — they’re how the product works by default, on every project.
Secrets stay encrypted
API keys and tokens are encrypted at rest. The agent can use a secret through controlled tooling, but the raw value is never printed back into chat or exposed to the model.
Guardrails on risky moves
By default, anything destructive or hard to undo is flagged for your approval instead of being pushed through quietly. Four permission modes — Plan, Ask before edits, Auto-accept, and Full autonomy — let you dial in how much the agent can do without checking in, and you can switch modes mid-conversation.
A clear audit trail
Every plan, tool call, and result stays visible in your project. You can always see exactly what the agent did, when, and why.
Checkpoints & rewind
Projects save restore points as you work, so you can roll back to an earlier state if an experiment goes sideways — without losing the rest of your progress.
You decide when to publish
Nothing leaves your workspace until you say so. Deploys and GitHub pushes are explicit actions you trigger — never automatic.
Our approach to your data
We collect the minimum we need to run your workspaces and your account. Project code and chat history live with your account so you can reopen work across sessions and devices. We do not sell your data, and we do not use your private project code to train models.
In transit and at rest
All traffic between your browser, our orchestrator, and your workspace is encrypted with modern TLS. Secrets such as API keys and tokens are encrypted at rest and decrypted only when needed to serve a request you initiated.
Isolation by design
Each project runs in its own isolated microVM with a dedicated filesystem and sandbox. The in-VM agent requires a per-project token, and VM-to-VM traffic across the shared egress bridge is dropped — so a bug, a runaway script, or a risky command in one project can’t reach another project’s machine. Mutable state is scoped to the VM that owns it.
Responsible disclosure
Found a vulnerability? We want to hear about it. Email security@gate15.dev with the details and steps to reproduce, and we’ll get back to you. We ask that you give us reasonable time to investigate and remediate before any public disclosure.
How it works
What happens to a request.
From your browser to the model and back, here’s the path your code and secrets take — and where the boundaries sit.
.env — the agent receives the env-var name, never the plaintext value, so secrets never enter the model context.FAQ
Security questions.
Who can see my code?
Your projects are private to your account and the teammates you invite. Each project runs in an isolated VM, and we don't use your private code to train models.
How are my model API keys handled?
Keys you add in Settings are encrypted at rest and used only to make requests on your behalf. They're never displayed back to you in full and never exposed to the agent.
Can I request a security review or DPA?
Yes. Enterprise customers can request a security questionnaire, a Data Processing Addendum, and a review of our architecture. Reach out through the Enterprise page.
What happens to a workspace when I close a project?
Workspaces pause when idle and resume exactly where you left off. You can delete a project at any time, which removes its sandbox and associated data.
Who are your sub-processors?
Anthropic, Z.ai, OpenAI, and Google process model requests; Vercel hosts the web app, and our own infrastructure runs the orchestrator and each project's isolated sandbox; Supabase is our database; and WorkOS handles authentication and SSO. We only use a provider when the feature you've turned on requires it — for example, your turn only reaches Z.ai, OpenAI, or Google if you select one of their models.
How is my data encrypted?
In transit, everything is TLS 1.3. At rest, project secrets and stored OAuth tokens (GitHub, Vercel, and the like) are encrypted with AES-256-GCM before they ever touch the database — the database never sees plaintext.
How long do you keep my data?
Your data lives for the life of the project and account so you can reopen work across sessions and devices. When you delete a project or account, its sandbox, secrets, audit events, and messages are hard-deleted via database cascade. To be candid: there is no time-based auto-purge today — data persists until you delete it.
How does the agent use a secret without seeing it?
Provider and connector secrets are decrypted only server-side, inside the orchestrator, and written directly into your sandbox's .env file. The AI agent receives just the env-var name and a confirmation — never the plaintext value. Your code reads it from process.env at runtime, so the secret never enters the model's context.
Is there an audit trail?
Yes. Project-scoped audit events are recorded for secret reads and writes, connector invocations, and checkpoint create/restore. A dedicated in-product audit viewer is not yet available; project activity remains visible in the workspace while the audit viewer is being built.
Are you SOC 2 or ISO 27001 certified?
Not yet — these certifications are on our roadmap, and we won't claim them until they're complete. In the meantime we're happy to walk enterprise teams through our architecture, isolation model, and controls, and to complete a security questionnaire and DPA.