What happened
Tilde.run introduced a platform enabling the safe deployment of AI agents in production environments. The system provides a versioned filesystem that integrates code from GitHub, data from S3, and documents from Google Drive into a single ~/sandbox. Each agent run operates within an isolated, serverless container, treated as a reversible transaction, allowing atomic commits of changes or instant rollbacks. Outbound network calls are audited and policy-checked, blocking unauthorised access to cloud metadata or private networks.
Why it matters
Production teams can now deploy AI agents with reduced operational risk, as Tilde's transactional sandboxes prevent unintended data modifications and exfiltration. This mechanism addresses the critical constraint of agent safety, allowing engineers to integrate agentic workflows into live systems without manual cleanup or backup restoration. This follows recent incidents, such as a Claude AI agent deleting a production database, highlighting the need for such containment.
