July

It’s been a landmark month.

Portkey is now the ✨official AI Gateway provider for Internet2 NET+✨, the gold standard for infrastructure across 300+ top U.S. universities including NYU, Harvard, Princeton, Cornell, and Berkeley. We’re now listed alongside AWS, Microsoft, and Google and we’re the youngest infra provider ever to make the cut!

Alongside this, we shipped major updates across the stack: expanded data residency support, enhanced logging and cost tracking for Claude Code, new providers and models, and more.

Let’s get into it!

Summary

Area	Key Highlights
Platform	• Regional data residency controls • Privacy mode • Automatic user attribution for user keys • Request transformations (original + final) • Log exports via UI, API, stream, webhook
Gateway & Providers	• New models: Grok-4, Qwen3-Coder, Kimi K2 • New providers: Krutrim, Agno AI • /v1/messages (beta): tool use, multi-turn, redacted thoughts • Custom models & pricing
Guardrails	• Palo Alto Networks AIRS plugin for real-time security guardrails • CRUD guardrails endpoint
Integrations	• LangSmith integration • AWS Bedrock Knowledge Bases • Langflow compatibility • Enhanced Claude Code logging & cost tracking

Internet2 🤝 Portkey

Many universities face similar challenges when adopting GenAI: fragmented provider access, budgetary uncertainty, and regulatory pressure. Portkey helps address these by sitting between institutional systems and model providers by offering a single API with granular usage controls, analytics, and guardrails.

“The Internet2 NET+ Service Evaluation process for Portkey has been a demonstration of collaboration and efficiency,” said Stratos Efstathiadis, Senior Director of Research Technology Services at NYU. “The synergy among peer institutions ensures Portkey meets the diverse AI requirements of higher education.”

📅 Join the NET+ Portkey Launch Webinar

NYU and Princeton leaders join us this Thursday, AUG 7 AT 1 P.M. ET to go behind the scenes of how GenAI is being deployed on campus. Register here

Platform

Regional data residency

All your LLM data can now be hosted in any region of your choice with strict isolation, no cross-border data flow, and full compliance with residency regulations.

• Meet institutional and regional privacy mandates

• Improve latency by routing traffic through nearby zones

Log exports

You can now export logs directly from the Portkey UI or through the API. Whether you’re debugging, auditing, or building internal dashboards, you can export and access complete logs in a structured, reliable format.

Transformed requests in logs

Debugging just got simpler. You can now view the original request as received by Portkey, the transformed version sent to the model provider, and same for the response in logs, making it easier for debugging.

Privacy mode (logging off)

Portkey now gives org owners granular control over what gets logged, from full request/response payloads to just minimal metadata.

This can be configured org-wide or per workspace, offering flexibility for sensitive or regulated use cases.Configure it for your org.

Automatic user attribution on API keys

Every request made using a user API key now automatically carries _user metadata, with an option to override.

Guardrails

Palo Alto Networks’ AIRS Plugin

Portkey now integrates with PANW AIRS (AI Runtime Security) to enforce guardrails that block risky prompts or model responses based on real-time security analysis. Learn more

CRUD Guardrails endpoint

We’ve added full support for managing guardrails via API — create, update, delete, and list guardrail rules and templates programmatically. Learn more

Gateway

Support for messages route(beta)

In addition to chat.completions, Portkey now supports the messages route (beta) for Anthropic, AWS Bedrock, and Vertex AI. This includes tool calling, thinking, multi-turn conversations, and redacted thinking support

New models and providers

• Krutrim: LLM provider with regional language support.
• Grok 4: The latest release from xAI’s Grok family, now supported on Portkey.

• Qwen3-Coder: Alibaba’s open-weight code model, available for direct use.
• Kimi K2: Moonshot’s new long-context model is now in production.

Integrations

Claude Code

Get cost tracking and usage breakdowns across teams for Claude Code generations — along with full request logs and output analysis to manage spend and performance.

AWS Bedrock Knowledge Bases

You can now connect to Bedrock Knowledge Bases directly through Portkey, with unified auth, caching, and logging across your RAG workflows.

Agno AI

Agno AI – Agno is a powerful framework for building autonomous agents, now supercharged with Portkey’s production-grade observability, governance, and tooling.

LangSmith

You can now export traces to LangSmith alongside Portkey’s built-in observability, for teams already using it for debugging or evals.

Langflow

Portkey now works natively with Langflow. Route all your workflows through Portkey to apply guardrails, monitor requests, and manage access without leaving your visual interface.

Portkey + your fav AI tool = 🪄✨

Resources

• Cookbook: Arize + Portkey: Multi-LLM debate with traces and evals
• Blog: How to add enterprise controls to OpenWebUI
• Blog: Everything We Know About Claude Code Limits

Community Contributors

A special thanks to our contributor this month:

• AG2AI-Admin
• Mishalabdullah

Coming this month!

MCP Connectors on the gateway! Reach out to us on [e-mail] for a sneakpeek!

Support

Need Help?

Open an issue on GitHub

---

Portkey is now PRISMA AIRS AI Gateway. See it in action.

Contact Us