Blekline
v0.3.0GitHub

Reference

Latency SLO

Enterprise ingress latency targets and measurement.

View as Markdown

Blekline ingress is not a sandbox runtime. Compare apples to apples:

LayerProductMetricTarget
RuntimeDaytonaSandbox create<80ms (vendor)
IngressBleklineTool-call policy (local)p99 <10ms
IngressBleklineFast-path mask (local)p95 <20ms
IngressBleklineAuthoritative mask (Azure)p95 <500ms

Fast-path modes

Set BLEKLINE_MASK_FAST_PATH:

ModeBehavior
azure_firstDefault — always call Azure after local prep
local_firstSkip Azure when local pass is clean (no names/PHI heuristic)
local_onlyNever call Azure — edge/regional agents

Per-request override: header x-blekline-mask-fast-path: local_first

Response headers

HeaderMeaning
x-blekline-latency-msTotal mask request time
x-blekline-mask-pathazure_first / local_first / local_only
x-blekline-ingress-regionEdge region label
x-blekline-mask-phasefast_local, completed, fallback_local

Benchmark

pnpm build:packages
pnpm demo:latency
# with token:
BLEKLINE_WORKSPACE_TOKEN=blw_... BLEKLINE_API_URL=http://localhost:3000 pnpm demo:latency

Health probe

curl http://localhost:3000/api/health/ingress

Edge sidecar:

curl http://localhost:8787/health

MCP proxy hot path

BLEKLINE_PROXY_LOCAL_ONLY=1

Cloud enforce runs async for audit; agent path stays local.


Next steps: AI Enablement Stack · Ingress proxy · Enterprise telemetry · Open workspace · Report issue