Portkey is an AI gateway and LLMOps platform for routing, monitoring, and securing LLM requests in production applications.
Teams connect through an SDK, and the vendor says setup takes three lines of code. Requests then pass through the gateway, where rules such as PII redaction and load balancing apply.
Palo Alto Networks completed its acquisition of Portkey, Inc. on May 29, 2026. The gateway is being integrated into the Prisma AIRS AI security platform. Portkey’s pricing page still lists free, Production, and Enterprise plans.
The platform targets developers and engineering leaders running LLM apps in production. Portkey states that it supports more than 1,600 models through one API.
Pricing
The free Developer plan includes 10,000 recorded logs per month and 3-day log retention. Logging stops past the cap, but requests are not blocked. The Production plan costs $49 per month and includes 100,000 logs, with overage at $9 per additional 100,000 requests up to 3 million. Enterprise pricing is custom, and a free self-hosted open-source version is also available.
* Disclaimer: Please note that pricing information may not be up to date. For the most accurate and current pricing details, refer to the official website.
Key Features
- ✓
Universal API covering 1,600+ models from many providers
- ✓
Automatic fallbacks, retries, and load balancing across providers
- ✓
Observability with 40+ metrics and OpenTelemetry-compatible dashboards
- ✓
Guardrails with 60+ checks, including JSON validation and PII anonymization
- ✓
Prompt studio with versioning, A/B testing, and instant deployment
Use Cases
Controlling High-Volume API Spend
A company running many LLM apps needs to stop unmonitored spending. Semantic caching avoids paying for repeated queries, and virtual keys enforce hard budget limits per team or user.
Keeping Support Chatbots Online
A customer support bot must stay available when its primary model fails or hits a rate limit. Fallbacks and retries reroute the failed prompt to a secondary provider.
Handling Sensitive Patient Data
A healthcare startup wants to summarize patient records with an LLM. PII anonymization guardrails strip identifying details from the request before it reaches the model.
Comparing Models on Real Documents
A product team must choose between GPT-4, Claude, and Mistral for its document format. The prompt studio runs side-by-side comparisons, so non-developers can test without code changes.
Blocking Prompt Injection Attempts
A public-facing AI app attracts users who try to override its instructions. Input guardrails from security partners screen manipulative prompts before the model processes them.
Strengths & Weaknesses
Strengths
One API reduces dependence on any single model provider.
A free, self-hostable open-source gateway suits privacy-conscious teams.
SDK integration into existing code needs only three lines, per the vendor.
Spend tracking works down to workspace, team, and individual user level.
Paid plans add semantic caching to cut cost and latency on repeated queries.
Weaknesses
The free plan keeps logs for only 3 days and allows 3 prompt templates.
Semantic caching and role-based access control require a paid plan.
Production overage stops at 3 million requests, so larger workloads need Enterprise.
Custom data residency, SSO, and BAAs are limited to Enterprise contracts.
Who Is This For?
AI application developers: They get traces and logs to debug multi-step agent workflows.
Engineering leaders: They gain fallbacks and load balancing to protect uptime and limit vendor lock-in.
Prompt engineers and product managers: They can test, compare, and deploy prompts without engineering help.
Security and compliance teams: They can apply guardrails and PII controls, with certifications on Enterprise.
Frequently Asked Questions
Who owns Portkey now?
Palo Alto Networks completed its acquisition of Portkey, Inc. on May 29, 2026. The gateway is being folded into Prisma AIRS.
How long does integration take?
The vendor says the SDK needs three lines of code. You install it through npm or pip and point requests at the gateway.
Can I self-host Portkey?
Yes, through the open-source version. It includes the universal API, fallbacks, routing, load balancing, and a basic dashboard.
How does Portkey help reduce LLM costs?
Semantic caching returns stored answers to similar prompts on paid plans. Virtual keys let you set budgets and rate limits per team.
Which plan includes compliance certifications?
The Enterprise plan lists SOC 2 Type 2, GDPR, HIPAA, and custom BAAs. Confirm current certifications with the vendor before sharing regulated data.
Does the gateway see my prompts and responses?
Yes, traffic passes through it. The PII anonymizer can strip sensitive fields first, but you should review log retention for your plan.
Can non-developers use the prompt tools?
The Prompt Engineering Studio supports versioning and parallel comparisons across models. Product teams can run tests there without code changes.
How does Portkey help debug AI agents?
It records the chronological sequence of an agent run, including tool calls. LangGraph and CrewAI are listed as supported frameworks.
Does Portkey support Model Context Protocol servers?
An MCP Gateway centralizes authentication, access, and observability for MCP servers. Details of its limits are not covered in the available sources.
Portkey routes requests to model providers such as OpenAI, Anthropic, and Mistral. It lists Azure, AWS, and GCP for private cloud deployments, and Okta for SSO on Enterprise. LangGraph and CrewAI are supported for tracing. Guardrail partners include Patronus, Pangea, Pillar, Lasso, Prompt Security, Bedrock, Promptfoo, Aporia, and Acuvity. As the parent company, Palo Alto Networks is integrating Portkey into Prisma AIRS.