Case Study
How Wix scaled Al-native work to 5,000 employees with Willow
Read More
AI Agent Security

Willow vs Palo Alto Networks Portkey

Palo Alto Portkey manages LLM routing, provider budgets, and MCP traffic. Willow governs employee AI tools with discovery, scoped access, approved capabilities, and runtime approvals.

TL;DR
  • Start with Willow when the priority is governing the capabilities employees use: local MCP configurations, skills, plugins, approved distribution and selected tool approvals.
  • Willow combines publication checks with approval before selected tool calls, giving security a decision before distributing a capability and before a sensitive action executes.
  • Consider Prisma AIRS AI Gateway, formerly Portkey, only if unified model routing, reliability, provider budgets and MCP or A2A gateway operations are required. Its MCP controls include authentication, tool access, runtime policies and traces.

‍

A company rolling out employee AI tools needs control over the capabilities people install and the business actions those capabilities perform. An engineering team running AI applications also needs model routing, retries, provider budgets and request traces. Those requirements can overlap at the MCP gateway, while still creating different administration and rollout work.

Willow connects employee AI discovery to device rules, organization skills and plugins, and managed tool use. It gives security a way to act on local capabilities, distribute approved replacements and require a human decision before selected MCP tools run. The administrator can keep the capability decision, the employee assignment and the live tool policy within that workflow.

Palo Alto Networks has completed its acquisition of Portkey. The product now presents itself as Prisma AIRS AI Gateway. It combines model gateway infrastructure with MCP and agent controls. This guide compares that gateway offering with Willow.

How Willow and Prisma AIRS AI Gateway compare

Willow vs Prisma AIRS AI Gateway

willow
Requirement Willow Prisma AIRS AI Gateway
Govern employee local capabilities Scan Agent discovery and policies for MCP servers, skills and plugins. Central registries and policy for onboarded MCP and agent servers.
Distribute reviewed material Group-assigned skills, published plugins and build-time guard checks. MCP server/tool registry and A2A agent registry.
Control tool execution Tool enablement, selected human approval and runtime guards. Tool-scoped access, PII policies and pre-execution blocking.
Investigate activity Handled tool calls, identity and guard outcomes in audit logs. Searchable MCP invocations and unified workflow traces.
Manage AI spending Client plugin conversation and supported model limits; per-user MCP rate limits. Workspace provider budgets, request or token rate limits, caching and model routing.
Choose deployment SaaS, hybrid runtime or full on-premises. Managed, private-cloud, VPC and self-hosted options.

The overlap includes real MCP governance. Prisma AIRS AI Gateway can authenticate users, scope server and tool access, inspect content and stop unauthorized tool calls. Willow governs the employee capability lifecycle around its gateway: local discovery, device enforcement, reviewed instructions and a managed path for employees to use approved tools.

Govern the capabilities employees already have

Willow’s AI Discovery inventories local MCP configurations, installed AI tools, skills and standing agent instruction files. A security team can identify material on an employee’s machine even when that material was not registered with the managed gateway. The useful output is a reviewable inventory of what an assistant can load and which connections it can attempt.

Device policies let admins allow, warn on or block discovered MCP servers, skills and plugins. Scoping rules by group, user or device supports a targeted rollout. An admin can publish a rule for a specific team while another team retains a necessary capability. The draft state keeps policy preparation separate from live enforcement.

Those policies require the Scan Agent on the device. Machines without it can still use direct unmanaged connectors. Managed capabilities routed through Willow follow the gateway’s own controls. Treat the endpoint rollout and gateway configuration as separate pieces of the deployment, with each piece responsible for the activity it can actually govern.

Prisma AIRS AI Gateway offers a central MCP registry with team-scoped server and tool access. Its agent gateway also provides a registry and proxy for A2A servers. These are useful managed infrastructure objects. When the purchase is driven by local employee configuration, include the actual MCP files, skills and plugins in the demonstration so registration and device governance are evaluated against the intended workflow. This comparison focuses specifically on Prisma AIRS AI Gateway. Palo Alto Networks also offers endpoint AI security capabilities through Koi, including local AI discovery and endpoint enforcement. These broader capabilities should be evaluated separately when comparing enterprise-wide AI governance solutions.

‍

Give teams reviewed skills and plugins

Willow’s skills inventory supports organization-owned skills, group assignments, GitHub imports and review states. Skills can be bundled into plugins. The benefit is a governed distribution process for instructions and supporting material: employees can receive a capability selected for their role, and administrators can review how that capability changes before expanding its use.

Build-time guards evaluate toolkit and plugin contents when they are published. They can block a package that contains disallowed instructions or sensitive content before it reaches employees. That control complements runtime inspection by moving part of the review to the material being distributed, rather than waiting for the first problematic invocation.

A useful evaluation separates three objects that may all be called a capability: a local skill file, a managed MCP tool and an A2A agent capability. Willow’s local skill and plugin workflow is explicit about the material distributed to employee clients. Portkey’s agent registry and proxy govern agent-server interactions. Ask each demonstration to use the exact object your employees will receive or invoke.

Keep sensitive MCP actions reviewable

Willow lets admins disable individual tools and require human approval before a selected tool runs. Disabled tools are hidden from clients using that gateway. A team can expose a read-only toolset and retain confirmation for operations such as writes or deletes. That gives the operator an action-specific control independent of whether the model considers the request routine.

Runtime guards inspect the tool name, arguments and context before a routed request reaches the MCP server, and can inspect the response before returning it. A configured rule can block the request or mask sensitive content. These decisions belong to the managed call path, so the upstream tool and authentication configuration remain part of the security design.

Prisma AIRS AI Gateway has its own runtime MCP policies, including PII redaction, content filtering and blocking unauthorized invocations before execution. Its MCP access controls can be scoped to an organization, workspace, team or individual user. Willow connects selected human approval to its employee capability workflow, so the administrator can manage the reviewed package, the assigned group and the sensitive action as related operating decisions.

Willow also applies configured Guard Hooks to supported prompt and file-tool events in Claude Code, Cursor and Codex. The hooks fail open on timeout or loss of connectivity. Gateway runtime guards are stateless. Scope the rollout to supported client events and include failure behavior when deciding which high-consequence operations need gateway enforcement or explicit approval.

Separate employee usage limits from model gateway budgets

Willow’s Usage Hooks plugin enforces conversation-length policies and supported model budgets before a client sends a prompt. Saving a policy without deploying the plugin does not enforce it. Cursor reports its selected model on each prompt; Claude Code reports it at session start, so a later model change can be missed. Codex prompts do not provide the model name for model-specific budget enforcement.

That mechanism suits a team governing employee coding-client usage. Willow also has per-user gateway tool-call rate limits, optionally filtered by group and MCP server. A tool-call rate limit is a different control from a provider token budget. Naming that unit in the purchase requirements avoids treating several cost controls as equivalent because all are described as limits.

Prisma AIRS AI Gateway’s provider integrations support workspace budget and rate limits, including cost-based or token-based budgets. Its model gateway provides fallbacks, retries, routing and caching. This is a material reason to select it when the engineering requirement is operating model access across applications and providers, rather than deploying an employee client usage policy.

Willow’s optional endpoint optimizer hooks can compress tool output and route prompts when routing is configured. Choose the control path required for the workload: an employee client with installed hooks, or a provider-facing gateway operating application requests. Specify the budget unit, intended routing behavior and failure handling alongside that choice so the operator can assess the actual deployment.

Use identity and logs to explain the governed action

Willow supports human OAuth and identity-provider exchange paths, as well as independently owned machine users. A machine user receives access through its groups and has a rotatable secret. The upstream credential mode determines whose authority reaches the external service, so the gateway identity and the service credential should be reviewed together for each MCP server.

Willow logs connect handled calls to the user, AI client, server, tool and guard decision. Prisma AIRS AI Gateway provides searchable MCP invocation logs and unified traces. For employee governance, start with the record needed to explain a tool decision. For application operations, include the cross-model and tool trace needed to diagnose a failing workflow.

Human SCIM lifecycle and non-human credentials remain distinct in Willow. Deprovisioning a human account does not automatically retire every machine-user or background-agent secret. Assign an owner and rotation or deactivation process to those identities. This is particularly relevant when a department turns an employee experiment into an unattended automation and changes the authentication path.

When Prisma AIRS AI Gateway is the right choice

Choose the Portkey-derived gateway only if the project requires its unified model infrastructure: provider-level workspace budgets, routing and fallbacks, or the operation of managed MCP and A2A agent endpoints alongside model traffic. These are concrete engineering requirements. For those workloads, provider credentials, model routes and agent endpoints are infrastructure the engineering team needs to configure, observe and maintain through its production operating process.

Its deployment choices include managed service, private cloud, VPC and self-hosting. Its public plans also separate open-source and hosted options from Enterprise features such as SSO and private-cloud deployment. Keep the edition and deployment scope visible in the comparison; an open-source gateway and a fully managed enterprise deployment should not be treated as one interchangeable package.

Deployment and pricing

Willow offers SaaS, hybrid runtime and full on-premises architecture. Hybrid retains the management and associated services in Willow SaaS, while full on-premises places the complete stack inside the customer environment. Both Willow and the Portkey-derived gateway have private deployment choices, so data residency alone should lead to an architecture comparison with the required services explicitly included.

As of October 21, 2026, Willow Enterprise is priced per human seat, billed annually with volume tiers. Portkey’s public pricing lists a free Developer plan and a $49-per-month Production plan with 100,000 recorded logs and $9 per additional 100,000 requests, while Enterprise is custom-priced. The $49 plan is not the price of an enterprise Prisma AIRS deployment. Request its current enterprise packaging and deployment quote.

Roll out reviewed capabilities with Willow

Willow gives the administrator a path from an employee’s existing local configuration to an approved capability. Review the discovered skill or MCP connection, publish the applicable device rule and assign the reviewed replacement to the intended group. The department receives a toolset selected for its work, while the organization retains the capability review and distribution decisions.

A plugin can carry skills, commands, toolkits and MCP servers into supported employee clients. Publication guards check that material before it is distributed. Selected-tool approval then supplies a separate human decision when the capability invokes a consequential action through Willow. These controls connect the package the employee receives with the managed tool call that follows.

For ongoing administration, keep a human owner on machine users, choose the upstream credential mode and review the handled-call records. Client usage policies and per-user MCP limits add controls with their own defined enforcement paths. This is a practical starting scope for employee AI governance. Provider routing, workspace budgets and A2A server operations become reasons to choose Prisma AIRS AI Gateway only when those infrastructure mechanisms are part of the project.

Which product fits the requirement

Which Product Fits the Requirement

willow
Main requirement Product to consider Reason
Govern employee local MCPs, skills and plugins Willow Discovery, device rules and reviewed capability distribution.
Require human approval before selected tool calls Willow Per-tool confirmation in the employee gateway workflow.
Operate provider budgets, routing and fallbacks Prisma AIRS AI Gatewayonly if required Provider-facing model controls and reliability mechanisms.
Manage A2A agent servers beside model and MCP traffic Prisma AIRS AI Gatewayonly if required Agent registry and server proxy in the gateway platform.

Choose Willow to govern the AI capabilities employees install, receive and invoke. Local policies, reviewed skills and plugins, publication checks and selected tool approval give the team a defined workflow for that responsibility. Explore Willow Enterprise to scope the employee rollout and deployment your organization needs.

FAQ

Which Palo Alto product does this Willow comparison cover?

Palo Alto Networks completed the acquisition, and Portkey now presents itself as Prisma AIRS AI Gateway. This comparison covers that gateway offering and its MCP and agent controls, rather than the entire Palo Alto Networks portfolio.

Why choose Willow when Prisma AIRS has MCP controls?

Willow connects local device policies and reviewed skills and plugins to selected human approval. Prisma AIRS also has MCP authentication, scoped tool access, runtime policies and traces. Choose Willow for governing the employee capability lifecycle; select Prisma AIRS only when its model or agent infrastructure is required.

Why choose Willow for employee AI governance?

Willow connects local capability discovery to device policies, approved skills and plugins, build-time checks and selected tool approvals. That combination supports the employee workflow from what is installed to what can be invoked through the managed gateway.

Are Willow usage limits equivalent to provider budgets?

No. Willow’s client plugin applies conversation and supported model limits before prompts are sent. Portkey provider integrations can apply workspace cost or token budgets and request limits. The control path, budget unit and client reporting requirements differ.

Does Willow support model-specific limits in Codex?

Codex prompts do not provide the model name, so Willow’s model-specific budget policy cannot stop those prompts. Conversation-length limits use the Usage Hooks plugin instead. Cursor and Claude Code have different model-reporting behavior.

Can Willow require approval before a tool executes?

Yes. An admin can require human confirmation for every call to a selected gateway tool. Runtime content guards can also block or mask content, allowing the team to configure approval and content policy as separate controls.

Can both products run in private infrastructure?

Both offer private deployment choices. Willow has hybrid and full on-premises architectures; Portkey lists private-cloud, VPC and self-hosted options. Specify the edition and the location of runtime, management and logs before comparing the proposals.

How should Willow and Prisma AIRS pricing be scoped?

Willow Enterprise is annual and per human seat. Portkey’s $49 public Production tier has a defined log allowance and overages; Enterprise is custom-priced. Its SSO and private-cloud scope belongs in an enterprise quote. Compare the required deployment and controls in each proposal.

‍

Table of contents

    Willow vs Maxim

    Maxim Bifrost focuses on LLM routing, caching, and developer-focused MCP gateway tools. Willow governs employee AI tools with shadow AI discovery, managed connectors, resource-level security checks, and built-in human approvals for sensitive actions.

    ‍

    Willow vs Microsoft Entra Agent ID

    Microsoft Entra Agent ID manages agent identities, Conditional Access, and lifecycle governance. Willow governs employee AI tools with MCP discovery, device policies, approved skills and plugins, and human approval for sensitive actions.

    ‍

    Willow vs JumpCloud

    JumpCloud manages identities, devices, SSO, and agent access through Agentic IAM. Willow governs employee AI tools with MCP discovery, device policies, approved skills and plugins, runtime security, and human approval for sensitive actions.

    ‍

    Your agents are already in the wild.

    Give them a Basecamp. Go from AI chaos to AI work, in minutes.