Security

Your Code Stays Yours

Agen agents run in isolated cloud sandboxes, work on separate branches, and never merge without your approval. Here's how we keep your code and data safe.

Principle

Isolated Cloud Sandboxes

Every agent session runs in its own isolated environment. No shared state between sessions. No shared filesystem, no shared memory, no cross-session access. When a session ends, the sandbox is destroyed.

Principle

Branch Protection

Agents never push to main. Every change goes to a feature branch. A PR is opened for human review. Nothing reaches your production branch without explicit approval from your team.

Principle

Human Approval Before Merge

The agent writes the code, fixes the pipeline, and delivers a preview. You decide whether to merge. The agent cannot merge on its own. Your team is always the final gate.

Principle

Budget Controls

Set spending limits per agent, per session, per scheduled run. The agent stops when it hits the limit. No runaway costs. No surprise bills. You control exactly how much compute each task can consume.

Session flow

What Happens Inside an Agent Session

01

Agent clones your repo into an isolated sandbox

A fresh, ephemeral environment spins up. The agent has read/write access to the repo inside this sandbox only. It cannot access other repos, other sessions, or your production environment.

02

Agent works on a feature branch

All changes are committed to a new branch. The agent cannot modify main, master, or any protected branch. Branch naming follows a consistent pattern for easy identification.

03

Pipeline runs in the sandbox

CI/CD runs inside the isolated environment. If the pipeline fails, the agent fixes it within the sandbox. Your production pipeline is never affected.

04

PR is opened for review

The agent opens a pull request with a clean diff, commit history, and a live preview. The PR is marked for human review.

05

You review and merge (or don't)

You check the diff. You test the preview. You merge — or you close the PR and re-assign with more context. The agent never merges autonomously.

06

Sandbox is destroyed

After the session ends, the isolated environment is torn down. No data persists between sessions.

Data handling

What We Access, What We Store

Repository Access

Agen clones your repository into an ephemeral sandbox to perform work. We access only the repos you explicitly connect. We do not store copies of your code after sessions end.

Credentials & Secrets

API keys, tokens, and environment variables you provide are encrypted at rest and used only during active sessions. They are never logged, never exposed in PRs, and never accessible to other sessions.

Session Logs

Agent activity logs are retained for debugging and audit purposes. Logs contain task descriptions, agent actions, and pipeline outputs — not your source code. You can request log deletion at any time.

No Training on Your Code

Your code is not used to train AI models. Your repositories, PRs, and session data remain private to your organization.

Compliance

Where We Are on Compliance

StandardStatus
SOC 2 Type IIIn progress
Data encryption at rest
Data encryption in transit (TLS)
SSO (SAML)Available on Scale plan
Role-based access controlAvailable on Pro and Scale plans
Audit loggingComing soon
GDPR compliance

We're actively pursuing SOC 2 Type II certification. If your organization has specific compliance requirements, reach out at security@agenhq.com and we'll discuss what we can provide.

Responsible AI

How We Think About AI and Code

Autonomous agents writing production code raises legitimate questions. Here's our position:

Human-in-the-loop by default.

Agents don't merge. They don't deploy. They produce PRs for human review. Autonomy applies to the work, not the decision to ship.

Transparent diffs.

Every change the agent makes is visible in the PR diff. Nothing is hidden. Nothing is obfuscated. You see exactly what changed and why.

Budget limits prevent runaway work.

Agents stop when they hit the spending limit. They don't keep running and billing. You set the boundaries.

Preview before merge.

Cross-Repository Live Previews let you see the running app before you make any decision. Verify behavior, not just code.

FAQ

Frequently Asked Questions

Everything teams ask before they connect production repos to Agen.

Q01

Can the agent access repos I haven't connected?

No. Agents can only access repos you explicitly connect to Agen. There is no automatic discovery or access to other repos in your GitHub/GitLab organization.

Q02

Can the agent merge PRs without my approval?

No. Agents open PRs for review. Merging requires explicit human action.

Q03

Where do agent sessions run?

In isolated cloud sandboxes. Each session gets its own ephemeral environment that is destroyed after the session ends.

Q04

Is my code used to train AI models?

No. Your code, repos, and session data are never used for model training.

Q05

Do you support SSO?

Yes, SAML-based SSO is available on the Scale plan.

Q06

Who should I contact for security questions?

Email security@agenhq.com. We'll respond within one business day.

Security

Secure by Design. Autonomous by Default.

Connect your repos with confidence.

$20 free credits · No credit card · Isolated sandboxes