Preloop vs Portkey: Agent control plane or AI gateway?

Preloop is the open-source AI agent control plane: an MCP firewall, AI model gateway, human approvals, runtime observability, and audit trail in one self-hostable platform. Portkey is an AI gateway and LLM ops platform (portkey.ai) that Palo Alto Networks acquired, closing on May 29, 2026, to serve as "the core AI Gateway for Prisma AIRS". The honest one-liner: if your problem is "route LLM traffic and run LLM ops for many applications," Portkey is a strong option, now inside a large security vendor; if your problem is "govern AI agents end to end, including MCP tool calls, approvals, and runtime sessions," choose Preloop.

At a glance

Feature Preloop Portkey
Ownership Independent Palo Alto Networks, part of Prisma AIRS
License Apache 2.0 (open source core) Commercial platform; the gateway is MIT open source
Self-hosted option Yes, the whole control plane on any infrastructure Open source gateway; private cloud and VPC hosting on Enterprise
Model gateway with budgets Yes: OpenAI-, Anthropic- and Gemini-compatible, per-agent attribution Yes: universal API, fallbacks, load balancing, retries; the gateway repository describes routing to "1,600+ LLMs" (GitHub)
MCP tool governance MCP firewall: ordered allow / deny / require-approval / require-justification rules with CEL on arguments MCP Gateway that centralizes authentication, access, and observability of MCP servers
Human approvals Per call: mobile, watch, Slack, Mattermost, email, webhook, CLI Not described on Portkey's homepage or pricing page at the time of review
Observability Session timelines: tool calls, model calls, decisions, approvals, spend Logs, traces, feedback, metadata, alerts for LLM calls
Guardrails Model content policies (PII, prompt-injection heuristics, moderation) on governed traffic Deterministic, LLM, and partner guardrails
Prompt management Not a focus Yes: templates, playground, versioning
Pricing Open source free; Cloud and Enterprise plans Developer free (10k logs per month), Production $49/month, Enterprise custom
Onboards existing local agents Yes: preloop agents discover and preloop agents onboard No direct equivalent

What Portkey does well

Portkey is a credible LLM ops platform:

  • A broad AI gateway. Many providers, retries, fallbacks, load balancing, caching, and guardrails at the gateway layer (pricing).
  • LLM observability and prompt tooling. Logs, traces, feedback, prompt templates, versioning, and a playground are first-class.
  • Scale. Palo Alto Networks describes its "ability to process trillions of tokens" (press release).
  • A security vendor behind it. For buyers standardizing on Palo Alto Networks, Portkey now arrives as part of Prisma AIRS.

If you are building many LLM-powered features and need a gateway plus prompt management, Portkey is a solid pick.

Where Preloop is different

Preloop and Portkey both include a model gateway, but Preloop is agent-first rather than LLM-app-first.

  • Policy on the tool call. Preloop's MCP firewall evaluates every governed tool call against ordered rules and decides whether to allow, deny, require approval, or require justification, with CEL conditions on the arguments.
  • Human approvals on critical actions. Preloop routes sensitive actions to mobile, watch, Slack, Mattermost, email, or webhook with full context. Async approval mode lets long reviews finish without breaking the agent's transport.
  • Native agent actions. Native action gates for Claude Code, Codex CLI, Cursor, Copilot CLI, OpenCode, and others put shell commands and file edits on the same approval path. No gateway sees those calls.
  • Runtime session view. One timeline per agent session shows every tool call, model call, matched policy, approver, spend, and outcome.
  • Independent and open source. The whole control plane, not only the gateway, is Apache 2.0 and self-hostable, and it is not part of a larger security suite.
  • Onboarding. preloop agents discover finds local Claude Code, Codex CLI, Cursor, Gemini CLI, OpenClaw, and OpenCode configurations; preloop agents onboard routes the supported ones through Preloop, so governance covers developer agents on laptops, not only server-side LLM applications.

When to choose Portkey

  • Your primary problem is an LLM gateway and LLM ops for many applications and providers.
  • You want rich prompt management and experimentation as a central feature.
  • You are standardizing on Palo Alto Networks and Prisma AIRS.

When to choose Preloop

  • Platform and DevEx teams rolling out AI coding agents who need one control plane for tool calls, model traffic, and approvals.
  • Security teams that need a self-hostable control plane, policy-as-code, and searchable audit evidence inside their own network.
  • Open-source-first organizations that need a control plane they can read, fork, and audit, from an independent vendor.
  • Teams building EU AI Act readiness evidence that need durable records of matched policies, approvers, and outcomes mapped to runtime sessions.

Adoption notes

Preloop and Portkey can coexist. A common pattern keeps Portkey as the LLM gateway for product workloads and uses Preloop as the control plane for developer-side agents and MCP workflows. Server-side agents can point at the Preloop gateway URL at the same time approvals and MCP policies are introduced.

FAQ

Is there an open-source alternative to Portkey? Preloop is an open-source AI agent control plane under Apache 2.0, with an AI model gateway, budgets, and attribution plus MCP tool governance and human approvals. For teams that need a broad LLM gateway only, LiteLLM and Portkey's own MIT gateway are open source options.

Who owns Portkey now? Palo Alto Networks, which closed the acquisition on May 29, 2026.

Does Preloop replace my AI gateway? For agent traffic, often yes: Preloop's model gateway is OpenAI-, Anthropic- and Gemini-compatible and supports many providers. Teams with prompt-ops workflows built on Portkey often add Preloop for MCP governance, approvals, and runtime observability and keep the existing gateway.

Can Preloop run self-hosted? Yes. The core is Apache 2.0 and deploys on your own infrastructure, in your VPC or on-prem.

Does Preloop support prompt management? Not as a product area. Teams with heavy prompt-ops needs pair a prompt-management tool with Preloop.

Try Preloop

Last reviewed: 2026-10-08.