Legal

Security

Last updated July 2026

Fulminare runs autonomous code on your behalf, so security is foundational. Here's how we protect your work and your data.

Isolated execution

Every chat runs in its own isolated cloud sandbox (E2B), never shared with another chat or another user. It's paused as soon as each turn ends and resumed by the next message, and a sandbox left untouched for 14 days is deleted. Nothing you run reaches another user's machine.

Data protection

  • Encryption in transit (TLS) and at rest.
  • Row-level security so you can only ever access your own sessions, memory, and tokens.
  • Integration tokens are stored encrypted and used only to fulfil your requests.
  • Google integrations request the narrowest scope that does the job (e.g. drive.file (access only to files Fulminare itself creates) rather than full account-wide access).

Rate limits & abuse prevention

Every plan carries hard, server-enforced limits: concurrent sandboxes, runs per day, a monthly token budget, and overnight-run allowances (see Terms for exact numbers). These are checked before a run starts, not after, so one account can't exhaust shared infrastructure or run up an unbounded bill. Sub-agent fan-out is similarly capped per run. If the usage check itself is unavailable, the run is allowed rather than blocked, so an outage on our side never locks a user out.

Human-in-the-loop

The agent surfaces every tool call as it happens, and you can stop a run at any moment. Overnight runs flag irreversible decisions rather than guessing.

Responsible disclosure

Found a vulnerability? We appreciate coordinated disclosure. Email security@fulmina.re and we'll respond promptly. Please don't access data that isn't yours while testing.