Deploy

Kubernetes fleet

Track 01 platform eval — mandatory-hop sidecar, Trust Vault, Lineage Firewall, and fleet policy in your cluster. Prove production-grade NHIM enforcement in ~2 hours.

View as Markdown

Eval tracks

Engineered for regulated verticals

01Runtime agent governance

NHIM lineage, runtime intent validation, and human-in-the-loop kill switches for AIUC-1 aligned programs.

02Threat mitigation

Defense against LLM01 prompt injection, LLM02 insecure output handling, and LLM06 sensitive information disclosure.

03Data sovereignty & audit

Stateful PII tokenization at the pod boundary and immutable execution logs for transparency workflows.

04AI management controls

Technical safeguards, audit logging, and data masking requirements for ISO 42001 certification audits.

05Risk management

Enforcement layer mapping to NIST AI RMF Govern, Map, Measure, and Protect functions.

Track 01 deploys Blekline NHIM runtime enforcement in your cluster — standalone sidecar (ClusterIP) or injected sidecar (mutating webhook). This is the primary platform eval path for regulated production teams.

Plan ~2 hours for a full eval pass with posture upload and first enforce events in Activity.

Step 0 — NHIM audit

Run static audit before sandbox install:

kubectl apply -f https://raw.githubusercontent.com/Blekline/blekline-oss/main/packages/nhim-audit/deploy/rbac/nhim-audit-reader-namespaced.yaml -n nhim-eval
kubectl apply -f https://raw.githubusercontent.com/Blekline/blekline-oss/main/packages/nhim-audit/deploy/rbac/nhim-audit-reader-cluster.yaml
npx @blekline/nhim-audit audit --profile generic --namespace nhim-eval --plain --json -o nhim-audit.json

→ [NHIM audit quickstart](/docs/get-started/nhim-audit-quickstart). Attach JSON to your sandbox grant for a scoped probe token.

Images

ghcr.io/blekline/sidecar:0.2.1-nhim
ghcr.io/blekline/admission:0.2.1-nhim

Topology A — Standalone sidecar

Platform team runs one sidecar Deployment; agent pods call it over ClusterIP.

kubectl create namespace blekline
kubectl -n blekline create secret generic blekline-sidecar-auth \
  --from-literal=token="$(openssl rand -hex 32)"
kubectl -n blekline create secret generic blekline-vault-master \
  --from-literal=masterKey="$(openssl rand -hex 32)"

helm upgrade --install sidecar blekline-ingress \
  -n blekline \
  -f sidecar/values-production.yaml

Agent env: BLEKLINE_SIDECAR_URL=http://<release>-blekline-ingress:8787

Never expose port 8787 on public Ingress.

Topology B — Injected sidecar

Mutating webhook adds blekline-sidecar to agent pods. Agents use http://127.0.0.1:8787.

Opt-in annotation:

metadata:
  annotations:
    blekline.com/inject-sidecar: enabled
helm upgrade --install blekline-admission blekline-admission \
  -n blekline --create-namespace

Mandatory-hop NetworkPolicy

Agent pods must route tool and model traffic through the sidecar — no bypass paths to the public internet or sensitive backends.

See sandbox pack for reference NetworkPolicy manifests and probe scenarios.

Auto-Route (Tier 1)

Auto-Route mode (three enforcement tiers)

When the mutating admission webhook injects blekline-sidecar, it also rewrites SDK env vars on agent containers so OpenAI / Anthropic clients route through http://127.0.0.1:8787 automatically.

TierMechanismCode change?
1 — Auto-RouteWebhook env inject → sidecar masks LLM ingress/egressNo (OpenAI/Anthropic SDK)
1b — iptables Auto-RouteInit container redirects HTTPS without env varsNo (Phase 2 — annotate blekline.com/auto-route: iptables)
2 — Mandatory hopNetworkPolicy → agent egress only via sidecarNo (network)
3 — Tool enforcePOST /v1/enforce-tool-call or @blekline/mcp-proxyYes* (MCP tool argument policy)

* Tier 3 is required for MCP tool argument policy (ASI02). Auto-Route covers model API paths only.

Injected env vars (Tier 1)

Env varValue
BLEKLINE_AUTO_ROUTEtrue
BLEKLINE_SIDECAR_URLhttp://127.0.0.1:8787
OPENAI_BASE_URLhttp://127.0.0.1:8787/v1
OPENAI_API_BASEhttp://127.0.0.1:8787/v1
ANTHROPIC_BASE_URLhttp://127.0.0.1:8787

Opt-out per pod: blekline.com/auto-route: disabled

Roadmap dataplane (not default shipped)

  • iptables init redirect — transparent HTTPS redirect without env vars (Tier 1b, Phase 2)
  • eBPF / Cilium — optional dataplane integration for platform teams (Phase 3)

See enterprise sandbox AUTO_ROUTE.md for operator install paths.

Apply mutating admission with blekline.com/inject-sidecar: enabled on agent workloads. OpenAI/Anthropic SDK env vars rewrite to 127.0.0.1:8787 — no application code change for LLM paths.

Verify eval success

  1. /health returns 200 with sidecar auth token
  2. First allow/mask/block event in Operations → Activity
  3. Posture JSON uploaded with score ≥ 70 (or CRITICAL gaps documented with probe plan)
  4. Optional: probe token validates runtime bypass resistance

→ [NHIM verification](/docs/enterprise/nhim-verification) · [Trust Vault sidecar](/docs/enterprise/trust-vault-sidecar) · Lineage enforcement

Enterprise sandbox

Track 01 sandbox packs include Helm values, mandatory-hop templates, checklist steps, and probe token workflow. Request via [platform eval signup](https://app.blekline.com/auth/signup?intent=platform) or book a pilot.


Other tracks: Docker sidecar · MCP self-serve · [Eval journey](/docs/get-started/eval-journey)