The Penaxtra Runtime AI Gateway is a self-hosted agent binary that proxies LLM API calls and applies DLP rules on the wire. Prompts never leave the customer VPC in this deployment mode.
Prerequisites
- A workspace with an
owneroradminrole - A host with outbound HTTPS access to the Penaxtra control plane
- A Linux amd64 or Windows amd64 host with at least 256 MB RAM and 100 MB free disk
Step 1: Mint an enrollment token
Sign in to the workspace and navigate to AI Runtime Gateway → Agents → Mint enroll token. Tokens are short-lived (15 minutes) and single-use. Copy the token to the host where the agent will run.
Step 2: Download the binary
The agent is shipped as a single cross-compiled binary. The latest signed release is available at /app/runtime/download from inside the workspace.
curl -fLo penaxtra-gateway \
-H "Authorization: Bearer $ENROLL_TOKEN" \
https://penaxtra.com/api/v2/runtime/binaries/linux-amd64
chmod +x penaxtra-gateway
Step 3: Verify the binary
The binary ships with a SHA-256 hash in the workspace download page and is signed at build time. Verify before running.
sha256sum penaxtra-gateway
# compare with the hash shown in the workspace download page
Step 4: Enroll + run
./penaxtra-gateway \
--enroll "$ENROLL_TOKEN" \
--listen 127.0.0.1:8443 \
--upstream https://api.upstream-llm-provider.example
The agent registers itself with the control plane, pulls the signed policy blob, and begins proxying. Point your LLM client at http://127.0.0.1:8443 instead of the upstream provider.
Security notes
- The enrollment token is consumed on first use; intercept it once and it cannot be replayed.
- The agent verifies the Ed25519 signature on every policy blob before applying it.
- Outbound TLS to the control plane uses TLS 1.3 with a pinned CA bundle.
- Prompt content never leaves the agent host. Only redacted block events and counters are forwarded to the control plane.
Related
- Kubernetes deployment
- Docker deployment
- Webhook signature verification
- Runtime AI Gateway platform overview
Need help? Email [[email protected]](mailto:[email protected]).
Last reviewed: 2026-06-13. Reviewed by: Engineering. Content type: Developer documentation. Reach the maintainers: [email protected] .