Kubernetes-native control plane
Define agents, routes, and policies through Kubernetes resources, Helm profiles, and standard service contracts.
Learn concepts ->
Governance for enterprise AI agents
Govern AI agents without slowing teams down.
AgentFleetGateway helps teams explore and scale AI agents without giving them unchecked access. Every request is governed by clear policies, protected model and tool boundaries, human approvals, and auditable controls, so teams can move quickly while staying in control.
Policy on every request
Human approvals
Protected model access
Console, CLI, and GitOps
Why it exists
Move from agent experiments to controlled operations.
Keep the frameworks your teams already use while adding a consistent control layer for identity, policy, routing, approvals, model access, and operations.
Governed execution
Apply routing, policy, approvals, and model-access controls before requests reach agents or external systems.
Explore operations ->Audit-ready visibility
Give operators a clear view of inventory, decisions, approvals, usage, cost, health, and request outcomes.
Explore operations ->
Architecture
One gateway, modular control-plane services.
Routing, policy, model access, tools, context, approvals, and audit remain separate modules, so teams can adopt the controls they need without replacing their agent frameworks.
Component map
Kubernetes APIs declare intent. Services enforce it at runtime.
GitOps, Helm, and CRDs
Agent
AgentRoute
AgentPolicy
Operator
Gateway
Registry
Scheduler
Policy
Model Router
Guard
Context
Approval
Workload Identity
MCP Router
A2A Adapter
Console / CLI
Audit, metrics, usage, cost, and traces
Request path
Agent selection and protected model access use separate governance boundaries.
User app
Gateway
Policy decision
Scheduler selection
Optional context
Agent service
Workload identity
Model Router
Optional Guard
Audit + usage evidence
MCP + A2A
Standards-first boundaries bring tools, model endpoints, and peer agents under control-plane policy.
MCP
Server registry
Tool discovery
Governed calls
Policy
AgentFleetGateway
Audit metadata
A2A
Agent Card preview
Import preview
Policy-controlled delegation
Bring model endpoints, MCP tools, and A2A peers under the same policy and audit boundaries. Review Compatibility for environment-specific requirements.
Get started
Start on managed Kubernetes.
Connect to an existing AKS, EKS, or GKE cluster, install with Helm, verify service readiness, then send your first governed request.
kubectl config current-context
kubectl apply -f crds/
tar -xzf agentfleet-<version>.tgz
helm upgrade --install agentfleet ./agentfleet \
--namespace agentfleet-system --create-namespace \
-f agentfleet/profiles/local-durable.yaml \
-f agentfleet-images.values.json
kubectl get pods -n agentfleet-system
Documentation
Documentation for platform teams.
Start with installation and architecture, then move into policy, security, operations, integrations, and reference material.
Overview
Understand the product, its scope, and the operator workflow it proves.
Architecture
See how gateway, registry, policy, protected model access, Guard, context, and evidence fit together.
Protected model path
Understand workload identity, controlled egress, credential isolation, inspection, and agent suspension.
Operations
Explore usage evidence, approvals, `agentfleetctl`, observability, and tenant-aware controls.
Compatibility
Review supported platforms, integration boundaries, and environment-specific requirements.
Enterprise early access
Evaluate AgentFleetGateway with your Kubernetes platform, identity provider, model endpoints, and governance requirements.
Request access ->