Open source · MIT
Oversight infrastructure for AI agents.
Stop your agents before they spend money, delete data, or send the wrong email. Add one decorator. Get human-in-the-loop approval, an immutable audit log, and oversight that actually works in production.
from sentinel import oversight
@oversight(risk_level="high", approvers=["sms:+15551234567"])
def transfer_funds(amount: float, to_account: str) -> dict:
"""Wire transfer. Requires human approval before executing."""
return banking_api.transfer(amount, to_account)Integration
Framework agnostic
Works with LangChain, LlamaIndex, CrewAI, AutoGen, or your own loop. Just decorate the function.
Trust
Hash-chained audit log
Every approval, decision, and execution is recorded with a cryptographic hash chain. Tamper-evident by default.
Control
Configurable fallbacks
Block on timeout, execute on timeout, or escalate. You decide what's safe for each function.
Pricing
Simple per-approval pricing.
Free
$0
- 100 approvals/mo
- Email approver channel
- Dashboard
- Audit log (30-day retention)
- Community support
Pro
$199/mo
- Unlimited approvals
- SMS approver channel
- Webhooks (HMAC-signed)
- Branded sender domain
- Audit CSV export
- 99.5% SLA
- Priority email support
Enterprise
Talk to us
- Everything in Pro
- SSO (roadmap)
- Custom retention
- DPA
- Dedicated support channel
- Security review w/ threat model + pentest (under NDA)