Portkey

Portkey

Portkey is a unified AI gateway and LLMOps control plane that routes, observes, governs, and secures LLM traffic across more than 1,600 models through one API. Portkey gives teams a single OpenAI-compatible endpoint with fallbacks, load balancing, retries, and caching, then layers on full observability — logs, traces, cost and latency metrics — plus prompt management and governance like role-based access control, budgets, and virtual keys. It adds guardrails for PII redaction and policy enforcement and an MCP gateway for secure tool access. The core gateway is open source with more than 10,000 GitHub stars, aimed at engineering teams running production LLM apps.

Production credibility: Portkey was founded in January 2023 by Rohit Agarwal (CEO) and Ayush Garg. It raised a $3M seed led by Lightspeed in 2023 and a $15M Series A led by Elevation Capital in February 2026. The company states more than 3,000 GenAI teams use it, with a 10K-plus-star open-source core and a 99.9% uptime SLA; those figures are its own. One item to note: the homepage now carries Prisma AIRS AI Gateway branding associated with Palo Alto Networks, so the branding relationship is worth watching, though Portkey remains a live, self-serve product under its own name.

Key Features

• A universal OpenAI-compatible API to more than 1,600 LLMs with unified key management
• Reliability routing: automatic fallbacks, load balancing, retries, and timeouts
• Observability: request logs, distributed traces, cost and latency metrics, and alerts
• A prompt-management studio with templating and versioning
• Guardrails: PII redaction and configurable input and output policy enforcement
• Governance: role-based access control, virtual keys, and per-team budgets
• Simple and semantic caching to cut cost and latency
• An MCP gateway for secure tool access, with an open-source core

Ideal Use Case

Portkey fits engineering and GenAI teams putting multi-model LLM apps into production who need one control plane for routing, cost, monitoring, and governance. It is a strong fit for enterprises requiring guardrails, role-based access control, and private-cloud deployment for compliance. It is less necessary for a hobby project on a single model, where a lightweight proxy library is enough.

How Portkey differentiates

Portkey combines gateway, observability, prompt management, and governance in one open-source-core control plane, versus point tools like LiteLLM for routing or Helicone and Langfuse for observability. Its enterprise depth — role-based access control, budgets, VPC deployment, and compliance certifications — differentiates it from lighter proxy libraries.

FAQ

Q: What is Portkey? A: Portkey is a unified AI gateway and LLMOps control plane. One OpenAI-compatible API routes across 1,600-plus models with fallbacks, plus observability, prompt management, guardrails, and governance.

Q: Is it open source? A: The core gateway is open source with more than 10,000 GitHub stars. Portkey also offers hosted and enterprise plans on top of the open-source core.

Q: How much does it cost? A: There is a free developer tier, a Production plan around $49 per month with observability, guardrails, and access control, and custom enterprise pricing for high volume and private deployment.

Q: How is it different from LiteLLM or Helicone? A: LiteLLM focuses on routing and Helicone and Langfuse on observability; Portkey combines gateway, observability, prompt management, and governance in one control plane with an open-source core.

Q: Who funds Portkey? A: It raised a $3M seed led by Lightspeed in 2023 and a $15M Series A led by Elevation Capital in 2026.

tl;dr

Portkey is a unified AI gateway and LLMOps control plane that routes, observes, governs, and secures LLM traffic across 1,600-plus models through one OpenAI-compatible API, with an open-source core over 10K GitHub stars. It has a free tier and a $49-per-month Production plan, and raised a $15M Series A led by Elevation Capital. Best for teams running multi-model LLM apps in production; watch its Prisma AIRS branding.