What PaloNexus is not

PaloNexus is the authorization and accountability layer between AI agents and the enterprise systems they act upon. It decides what an agent is authorized to do and whose authority it is using — and it deliberately does not compete in the categories below. Here is the honest boundary, one category at a time.

Agent runtimes decide how an agent works. Sandboxes decide where its code runs. PaloNexus decides what it is authorized to do and whose authority it is using.

Not an agent framework

Agent frameworks — LangChain, LangGraph, Deep Agents, the OpenAI Agents SDK, CrewAI, AutoGen — own the agent loop: prompting, tool routing, memory, handoffs, and orchestration. PaloNexus does not build agents and ships no framework of its own. Instead it drops governance into the framework already in use: the shipped LangChain, LangGraph, and Deep Agents adapters gate tool, model, and agent-to-agent calls through the same /authz decision without restructuring the agent. See Keep secrets outside Deep Agents sandboxes.

Not an agent runtime

Runtimes and agent platforms — kagent is the clearest Kubernetes-native example — own building, deploying, connecting, running, and observing agents as workloads. PaloNexus does not run agents. It governs the authority those agents exercise, whichever runtime hosts them: the runtime keeps orchestration, memory, models, and observability; PaloNexus supplies accountable ownership, delegation validation, and short-lived scoped access. A kagent integration is planned — see Govern kagent agents with PaloNexus.

Not a sandbox provider

Sandbox products — Kubernetes Agent Sandbox, OpenAI’s sandbox workspaces, and the provider-neutral backends behind LangChain Deep Agents (E2B, Daytona, Modal, Runloop, and others) — own isolated code execution: filesystems, command execution, snapshots, and workspace lifecycle. PaloNexus does not provision or operate sandboxes. Its role starts exactly where isolation stops: once a sandboxed agent needs to touch a real enterprise system, PaloNexus is the trusted authority broker outside the sandbox, so the sandbox never holds standing credentials. See the planned Agent Sandbox and OpenAI Agents integrations, and the working Deep Agents pattern.

Not a model gateway

Model gateways and LLM routers own provider routing, model fallback, caching, and cost optimization across model vendors. PaloNexus does not route or resell model traffic as a product. It treats a model call as one more governed resource type: the same deny-by-default /authz decision, allowlist, budget, and audit record that apply to a tool call or an agent-to-agent hop apply to a model call. An existing model gateway can stay in place — PaloNexus sits on the authorization edge, not the inference path. See Credential-safe egress enforcement.

Not a Kubernetes operator

Kubernetes-native agent operations — declarative agent Custom Resource Definitions (CRDs), sandbox lifecycle CRDs, warm pools, node-level isolation — are owned by projects like kagent and Agent Sandbox, backed by cloud-native communities. PaloNexus deploys on Kubernetes and ships an integration with Envoy’s external-authorization hook (ext_authz) and network-layer egress enforcement as one enforcement adapter, but Kubernetes is a deployment target, not the product’s identity. The same authorization decision is reachable from an SDK adapter with no cluster at all. See Credential-safe action enforcement for the Kubernetes mechanics.

Not a workforce-IdP replacement

The workforce identity provider (IdP) — Okta, Entra ID, Google Workspace, Keycloak, Logto — remains the source of truth for humans: who exists, what role they hold, whether they are still employed. PaloNexus never re-invents human identity. It connects to the IdP over OpenID Connect (OIDC) and SCIM (System for Cross-domain Identity Management), and extends it downward to agents: every agent is bound to an accountable human owner from that directory, and joiner/mover/leaver changes in the IdP cascade into agent access and delegation revocation. See IdP support model and Connect agents to enterprise authority.

So what is it?

PaloNexus makes sure an AI agent can act only with authority that a real person or service owner was entitled to delegate — and only for the task, resource, and time originally approved. Start with the Overview, then browse the Integrations pages to see how it complements each ecosystem above.