Security
Agents act on your code and read your analytics. We assume an agent can be wrong, or be misled by what it reads, so each run can reach only what its job needs.
Every run is isolated
- Each agent run starts in a fresh, isolated environment of its own. It shares nothing with other runs or other customers, and it is deleted when the run ends.
- The credentials a run uses are kept apart from the agent itself, so the agent can't read them.
- A run can reach only the repository it works on, only for a short time, and only with the access its job needs.
- Runs have time limits, and a workspace can run only as many at a time as its plan allows.
Keys the agents never see
- Keys you enter are write-only: they are encrypted the moment they arrive, and nobody can read them back on the website. A short fingerprint tells keys apart.
- Each key is encrypted for its own workspace and purpose, so it can't be moved or reused anywhere else. Only the part of the platform that starts runs can decrypt keys.
- Your AI key never reaches the agent. The platform adds it on the way to your AI provider, and counts the tokens each run uses.
- Changing or removing a key needs your password, typed recently, and every change is in the audit log.
Integrations are read-only
- Agents read your analytics and error-tracking tools through the platform, never with your keys.
- They are offered only tools that read. Tools that would create, change or delete data are hidden and refused.
- Every call is logged: which integration and tool, whether it worked or was refused, and how long it took. What it read is not stored.
Guarded changes to your code
- An agent's changes are checked before anything reaches your repository: which files they touch, and how large they are.
- Agents can never change your CI and deployment setup, database migrations, dependencies, secrets and environment files, or their own instructions. Each project can block more paths and set its own size limit.
- Agents push only to their own branches and open pull requests. They never push to your default branch.
- A second model reviews every agent pull request before you're told it's ready to merge.
People stay in charge
- Plans, code and merges start only when a member of your workspace approves them.
- A merge needs your password, typed within the last 12 hours, so a stolen session alone can't ship anything.
- Sign-ins are protected against stolen sessions and requests from other sites, and repeated wrong passwords lock the sender out for 15 minutes.
- The audit log records every sign-in, approval, merge, key change and setting change: who, when and from where.
Your data
- Your data is stored and processed in the United States (Virginia), and is encrypted in transit and at rest.
- We never train models on your code or data. What agents send your AI provider falls under that provider's data policy for your account.
- Your code is checked out only inside a run's isolated environment, and is gone when the run ends. Reports, plans, findings and run records stay until you delete them or close your workspace.
- The privacy policy and the list of sub-processors name the providers we use and what each one receives.
More detail for your security team
This page describes what the platform guarantees, not how it is built. If your security team wants the details, such as the architecture, how keys are protected, or answers to a security questionnaire, e-mail contact@infrastudio.io and we'll go through them together.
Reporting a security issue
If you believe you have found a vulnerability, e-mail contact@infrastudio.io with the details. Please give us a reasonable time to fix it before telling others. We reply within two working days.
Questions from your security team?
We're happy to go through the details with them and answer a security questionnaire.