Portkey

Portkey has joined Palo Alto Networks. The enterprise AI stack has a structural problem. Teams are deploying autonomous agents that make high-stakes decisions across internal systems, at machine speed, with no centralized visibility or control. Legacy tools weren't built for this. Portkey was built to solve this. One gateway that handles routing, reliability, observability, and cost control across every model, agent, and tool your stack calls. No tradeoff between developer speed and security governance. This infrastructure now processes trillions of tokens a month, supporting enterprise teams running mission-critical workloads through Portkey. Joining Palo Alto Networks helps take this mission further. Portkey will become the AI Gateway for Prisma AIRS, the control plane for every AI transaction in the enterprise. Following the integration, this will unlock: • Runtime security and threat detection on agentic traffic • AI identity controls and least-privilege access for agents • Compliance across all AI workflows • Unified access to 3,000+ LLMs and MCP tools • Production reliability through automated failovers and more Enterprises shouldn't have to choose between moving fast and staying secure. That tradeoff is now off the table. To every team that ran production workloads through Portkey: you can expect the same gateway, same APIs, and same team. And, a lot more is coming. Read the full press release: https://lnkd.in/dUk9Uh3Y

New Opus BhaAi is live on Portkey. That’s it, that’s the post.

AI models move fast. New models drop constantly and your users expect access to the best ones without waiting on your team to catch up. See how Fontys ICT gets new models in front of staff and students on day zero

Giving everyone personal access to AI tools sounds like a win until no one has visibility into how it's being used, what it's costing, or whether it's compliant. Fontys ICT moved from unmanaged, personal AI access to institution-wide oversight with Portkey as the central layer. See how:

If you're rolling out AI across your organization, the infrastructure decisions you make early will define everything that follows. Fontys ICT built their AI platform with a few principles right from the start 👇

LLMs are only as good as the information they have access to. Tavily's real-time web search natively is now available in Portkey's gateway layer, so every prompt is automatically enriched with current, relevant information before it reaches the model. This allows teams to:  - Ground LLM responses in live web results without changing application code - Inject up-to-date context across agentic workflows, MCP calls, and tool invocations - Get full observability into what context was added, for every request - Configure once and scale it across every application routing through the gateway, across 1,600+ models

When an AI provider starts failing, you don't want your app to keep hammering it. Circuit breakers automatically detect this and stop routing traffic to that provider until it recovers. But circuit breakers need to know which provider to watch. If that reference shifts, the rule breaks silently and your protection disappears. Portkey's circuit breaker now resolves targets using stable provider slug references. Rename a provider, restructure your catalog, share configs across teams, and your breaker rules stay intact.

Who can call which model, and with what parameters, usually ends up as checks scattered across different services and clients. Control it centrally with our new guardrail checks: Model Rules — tie model access to request metadata. Allow/deny lives at the gateway instead of being re-implemented in every app. Request-params check — validate tools and parameters before the request reaches the LLM.

Running multiple providers and models in production requires more routing control. Here are a few we just added to Portkey: Passthrough targets: if you don't want the gateway to rewrite a specific hop, you can now send it straight to the upstream you name. Sticky sessions: follow-up calls in a conversation will land on the same backend or model instance instead of scattering. Forward headers: copy or rename inbound headers on the way out. Useful if you're passing tracing IDs, tenant markers, or provider-specific headers. Subdomain routing: if your routing logic depends on subdomain variants of your gateway host, Configs now respects that. Read more about this here: https://lnkd.in/gmUkCuiz

Your team probably uses Claude Desktop. But is it governed? You can now route Claude Desktop through Portkey and get: → Route to Anthropic, AWS Bedrock, or Vertex AI from a single endpoint → Per-team budgets, rate limits, and guardrails enforced automatically → Full logs — cost, latency, tokens, metadata — for every prompt Same Claude Desktop experience. See the setup guide here → https://lnkd.in/gai6fqWU