/ trust & security

Security is the baseline, not a feature.

Most platforms treat security as an add-on. We treat it as the only acceptable starting position. This page documents what's true today, what's on audit, and what's planned.

Architecture for security

Zero-trust networking

  • All service-to-service communication is mutually authenticated (mTLS) and encrypted.
  • Identities are issued by SPIFFE; no IP-based allowlists.
  • Layer-3-to-7 network policies enforced at the kernel via Cilium eBPF.

Runtime defense

  • Tetragon (eBPF) monitors every kernel syscall in production workloads.
  • If an application attempts a prohibited operation — shell escape, restricted memory access — the signal is killed in the kernel before execution.
  • All denials are auditable.

Supply chain

  • Every container image is signed by Cosign at build time.
  • Talos nodes are configured to refuse unsigned images via admission control.
  • Builds run inside Kata-isolated, rootless BuildKit microVMs — hostile by assumption.

Hard tenancy

  • On bare metal, every tenant runs inside a Cloud Hypervisor microVM. Hardware-level isolation, no shared kernel.
  • On public cloud, we use hardened native containers with user namespaces and Tetragon enforcement — provider-grade isolation.

Encryption

TypeStatusDetail
In transit (customer endpoints)ShippedTLS 1.3 default. Let's Encrypt for public services.
Service-to-service (in-cluster)ShippedWireGuard tunnels via Cilium ClusterMesh.
At rest (block storage)Roadmap (post-MVP)LUKS on Avahana metal; cloud-vendor SSE on cloud fleet.
Customer-managed keys (CMEK / BYOK)Roadmap (Sovereign tier first)HashiCorp Vault integration; key rotation managed by customer.

Compliance roadmap

DPDP Act (India)By GA
GDPR (EU)By GA — DPA already available
SOC 2 Type IIAudit post public beta
ISO 27001After SOC 2
HIPAAOn request — Sovereign tier
PCI DSSOut of scope (use a payment processor)

Incident response & SLA

  • Status page: status.avahana.ai (provisioning).
  • All P0/P1 incidents trigger an automatic public post-mortem within 7 days. No spin.
  • Customer-impacting events surface in-product and via email.
  • SLA targets are tier-specific and published in each tier's contract.

Reach security

  • Vulnerability disclosure: security@avahana.ai (GPG key on request).
  • We do not run a paid bug bounty yet. We do answer disclosures within one business day.
  • We will not threaten or pursue researchers who follow good-faith disclosure.