For security teams
Agents with limits you can audit, and keys that stay put.
Every agent holds a revocable virtual key instead of your provider key. Every call is checked and logged. Self-hosted, prompts and keys never leave your network.
200 · passed through flocktab.com, nothing stored · $0.04 settled
200 · straight to the vendor · only the numbers went to flocktab.com
What flocktab.com sees in each mode: hosted, the prompt passes through memory; self-hosted, only model, tokens and cents.
What agents change for security
- Provider keys get copied into every agent's environment, with no way to scope or revoke one agent.
- Agents run shell commands, ssh to other hosts and call payment APIs on their own judgement.
- Gateways that meter spend usually see and store every prompt.
With FlockTab
Virtual keys, not provider keys
Each agent gets an ft_live_ key, stored hashed and revocable instantly. A leaked key can spend at most what is left on its tab.
Actions judged before they run
Jev's guard stops known irreversible commands by rule and judges the rest, and blocks calls that reach past the project: ssh, sudo, reading keys, turning the sandbox off.
A clear data boundary
Self-hosted, nothing sensitive leaves. Hosted, keys are AES-256-GCM ciphertext opened per call with an unlock we never store; prompts are never stored.
Audit and identity
Every decision is logged with its reason. Fleet adds SSO, SCIM, roles checked on every request and an audit log of who changed what.
- Fails closed
- No prompt storage
- Passkeys and two-factor free
- Honest threat model on our security page
Review the data boundary.
Read the security page, then run the self-hosted proxy and see what leaves your machine.