Curate-Me vs Portkey
Portkey was an AI gateway for teams building production-grade AI apps. In April 2026, Portkey was acquired by Palo Alto Networks (~$120-140M valuation, double their February 2026 valuation) — it now serves as the AI Gateway for Prisma AIRS, Palo Alto’s AI security platform. Prior to the acquisition, Portkey open-sourced their entire gateway in March 2026, making the proxy layer a commodity.
Both platforms provide LLM routing, cost tracking, and observability. The difference is what happens beyond the proxy — and whether the roadmap is independent or tied to a security incumbent’s product strategy.
Feature comparison
| Capability | Curate-Me | Portkey |
|---|---|---|
| LLM proxy / routing | 50 providers (44 production-ready) | 1,600+ providers |
| Cost tracking and budgets | Per-request, per-key, per-org hierarchies | Per-workspace tracking |
| Rate limiting | IETF-standard headers, per-key RPM | Per-workspace limits |
| PII scanning | 33 regex patterns + optional Presidio NER | Content filtering (guardrails add-on) |
| HITL approval queues | Built-in, configurable by cost/risk | Not available |
| Managed execution environments | OpenClaw runners with lifecycle management | Not available |
| In-container governance | MCP server (50 tools) inside every runner | Not available |
| Desktop streaming | Live VNC/RDP agent monitoring | Not available |
| Agent-to-agent communication | A2A protocol with PII scanning | Not available |
| Time-travel debugging | Replayable execution audit trail | Log search |
| Self-hosted option | No (managed only) | Yes (MIT-licensed gateway) |
| Prompt management | Not available | Built-in versioning and A/B testing |
| Semantic caching | Response caching | Semantic + exact match |
Where Portkey is stronger
- Provider breadth: 1,600+ integrations vs our 50+. If you need a niche provider, Portkey likely supports it.
- Open-source gateway: The gateway code is MIT-licensed and self-hostable.
- Prompt management: Built-in prompt versioning and A/B testing that we do not offer.
- Established community: 6K GitHub stars and an active contributor base.
Where Curate-Me is stronger
- Managed runners: We provide sandboxed execution environments for AI agents, not just a proxy for LLM calls. Portkey has no execution capability.
- Human-in-the-loop: Our HITL gate flags high-cost or sensitive requests for human approval before they execute. Required for EU AI Act Article 14 compliance.
- In-container governance: Our MCP server runs inside every container, governing agent behavior from within. Portkey governance stops at the API boundary.
- Desktop streaming: Watch agents work in real-time via VNC/RDP.
- A2A protocol: Agent-to-agent communication with PII scanning and audit trail.
When to choose Portkey
- You only need an LLM proxy (no agents, no execution environments)
- You need 1,600+ provider integrations
- You want to self-host the entire stack
- Prompt management and caching are your primary concern
When to choose Curate-Me
- You run AI agents that need execution environments
- You need governance beyond the proxy layer (HITL, in-container controls)
- You need an immutable audit trail for regulatory compliance
- You want one platform instead of stitching Portkey + E2B + an observability tool
Ready to try it? Follow the Quick Start to connect in under 5 minutes.