> ## Documentation Index
> Fetch the complete documentation index at: https://docs.twill.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Security

> How Twill protects your code and data

Twill is built with enterprise security requirements in mind. Your code and credentials stay protected at every step.

## Code Execution Isolation

Agent code runs in isolated cloud dev environments, completely separate from Twill's infrastructure:

* **Per-workspace isolation**: Each workspace gets its own dev environment. One workspace's code, repos, and secrets are never reachable from another's.
* **Per-task forks**: Every task runs in an isolated fork, so changes made by the agent can't leak into your base environment or into other tasks.
* **Resource limits**: CPU, memory, and disk quotas prevent runaway processes.

See [Dev Environment](/sandbox-environment) for how the workspace sandbox and task forks fit together.

## Read-Only by Default

Twill agents operate with minimal permissions:

* **GitHub read-only**: Agents cannot push directly to your repositories
* **PR-based workflow**: All changes go through pull requests for human review
* **No direct deploys**: Twill never deploys code without your approval

## Data Encryption

Your sensitive data is protected at rest and in transit:

* **AES-256-GCM encryption**: Environment secrets are encrypted before storage
* **HTTPS everywhere**: All API communication uses TLS
* **Short-lived tokens**: GitHub access tokens are cached briefly and refreshed automatically

## Webhook Verification

All incoming webhooks are cryptographically verified:

* **HMAC-SHA256 signatures**: GitHub, Linear, and Slack webhooks use signed payloads
* **Timing-safe comparison**: Prevents timing attacks on signature verification
* **Replay protection**: Slack webhooks reject requests older than 5 minutes

## Workspace Isolation

Multi-tenant architecture ensures strict boundaries:

* **Role-based access**: Owner, Admin, and Member roles with appropriate permissions
* **Credential scoping**: Each workspace's secrets are isolated from others
* **OAuth state validation**: CSRF protection on all authorization flows

## No Model Training

Your code and prompts are never used to train AI models. Twill processes your data only to complete tasks you request.

## Best Practices

### Environment Variables

* **Use Twill's secret storage**: Add sensitive values through the workspace environment settings rather than committing them to your repository
* **Use dev credentials only**: Create separate API keys for development—never give agents access to production credentials
* **Rotate credentials regularly**: Update API keys and tokens periodically
* **Scope permissions narrowly**: Give third-party API keys only the permissions they need

### Branch Protection

Enable GitHub branch protection on your main branches:

* **Require pull request reviews**: Ensure human review before merging agent PRs
* **Require status checks**: Block merges until CI passes
* **Restrict direct pushes**: Prevent bypassing the PR workflow

These settings work with Twill's PR-based workflow to maintain code quality and security.

## Third-Party Services

Twill integrates with trusted infrastructure providers:

* **Modal / Daytona**: Sandbox execution environments
* **GitHub / Linear / Slack**: Issue tracking and communication
* **Stripe**: Payment processing (PCI compliant)

For detailed privacy information, see our [Privacy Policy](https://twill.ai/privacy).