Kanhe · EnterpriseXD-E · 2026

One gateway for every model.

Add a detection and routing layer in front of your models. Business code stays untouched, and there is only one configuration file.

The online demo needs no installation — try it directly in the browser.

TopYAML configurationOne file for every model · hot reload
MiddleGateway clusterEngine instances × N · shared prototype library
BottomModel backendsPublic cloud · private · hybrid

Docker / K8s deployment · YAML hot reload · multi-model routing

Architecture

Three-layer security architecture

Configuration, detection and routing each stay in their own place.

Top

YAML configuration

One configuration file governs every model and route. Changes take effect immediately — no restart, no business code changes.
Middle

Gateway cluster

Engine instances scale horizontally and share the prototype library and audit log. A single unhealthy instance does not break decision consistency.
Bottom

Model backends

Public cloud, private and hybrid deployments connect the same way. Models can be added or removed at any time without touching the decision policy.
Routing

Intelligent multi-model routing

Pick a model from a header or a field. Zero changes to business code.

Request arrives → parse X-Model-ID → look up the YAML routing table → land on the target model. The table supports weighting and failover, and changes need no restart.

# Enterprise multi-model gateway configuration
models:
  - id: deepseek-v3
    endpoint: http://internal.deepseek:8000/v1
    type: private
  - id: gpt-4
    endpoint: https://api.openai.com/v1
    type: public
    api_key: ${OPENAI_API_KEY}
routing:
  strategy: weighted
  fallback: deepseek-v3
Deployment

Pick a deployment by scale

From a single machine to a cluster, with one configuration file.

S

Docker Compose

Bring up the full gateway with one command. Good for PoC and small teams.

docker-compose up
M

Kubernetes Helm

One-command Helm deployment, autoscaling, shared prototype library across instances.

helm install kanhe
L

Cloud native

Managed runtime, scale on demand, fits ECS / EKS.

cloud deployment
Scenarios

Three everyday operations

Launch, scale, audit — each has a definite action.

New model goes liveAdding a new LLMYAML hot reload, protected immediately30 s
Traffic spikePeak QPS surgeScale engine instances horizontallyminutes
Security auditMonthly compliance reportAudit log on disk, exportableinstant
Rollout

From PoC to full coverage

Incremental deployment, with an acceptance point at every step.

Phase 1
PoC validation
1–2 weeks

Single model, Docker deployment

Phase 2
Multi-model
2–4 weeks

3+ models and routing working

Phase 3
High availability
1–2 months

K8s cluster and failover

Phase 4
Policy tuning
Ongoing

Lower false positives, accumulate rules

Covers four roles — security, operations, development and compliance — each seeing only their own view.

Start with one model

One demo is enough to see where it blocks and what it lets through.