An out-of-process governance layer for agent egress.
Panopticore runs in its own process and trust boundary, apart from the agent runtime, so the agent cannot alter its decisions or its evidence.
Three components, deployed in your VPC.
Edge Authenticator
- mTLS identity at the boundary
- Principal extraction from URI SAN
- CRL checking
- DSSE token minting
Sidecar Proxy
- Token verification
- Policy evaluation against OPA/Rego bundles
- Egress guardrails: allowlists, DNS pinning, rate limits
- Event capture to tamper-evident ledger
Admin Service
- Policy management
- Approval orchestration
- Slack integration with TOCTOU-safe signed tokens
- Evidence Binder rendering
Outside the agent's process and trust boundary.
Panopticore is an out-of-process control point between the agent and the external systems it calls. Nothing runs inside the agent's runtime, so the agent cannot modify, bypass, or disable it.
Security teams already use this pattern for external authentication services, network firewalls, and reverse proxies.
What the platform produces.
Policy decisions
Allow, warn, block, or approval-required. Deterministic. Evaluated against OPA/Rego policy bundles.
Approval workflows
Slack today, additional channels on the roadmap. Signed tokens prevent TOCTOU races between approval and execution.
Tamper-evident ledger
Every governance event is written to a signed, Merkle-linked ledger that auditors can check.
Evidence Binders
Cryptographically signed, self-contained session records that any third party can verify offline.
Learn more →Panopticore fails closed.
Identity validation failure
If mTLS identity cannot be established or the certificate fails CRL checks, the edge fails readiness and blocks all traffic.
Key rotation
Evidence remains verifiable under key rotation. Binders carry the verification artifacts needed to validate them against the key that was active at signing time.
Policy bundle updates
Policy bundles are versioned. The proxy evaluates against the bundle that was active at request time. Bundle checksums are recorded in the Evidence Binder so auditors can verify which policy was in effect.
Shipped and planned.
Further reading
AI systems cannot reliably govern themselves. Why deterministic, infrastructure-layer governance beats self-reporting by the agent under review.
Pre-deployment review, model evaluation, and prompt engineering all happen before the agent acts. Runtime governance happens when it acts.
See it in your environment.
Request early access and we'll scope a design partner engagement for your stack.