Portkey AI is a production AI gateway and control plane for teams routing LLM requests at scale. In March 2026 the entire gateway went fully open source. It processes 1T+ tokens/day across 24,000+ organizations, routes to 1,600+ models, and added a dedicated MCP Gateway for governing agent tool calls.
Why It's in Trial
Portkey went fully open source in March 2026 — every enterprise feature previously behind a SaaS paywall (governance, observability, auth, cost controls, MCP gateway) is now available self-hosted. That removes a key adoption barrier and puts it head-to-head with LiteLLM on the open-source gateway tier:
1,600+ LLM routing: Single OpenAI-compatible API across OpenAI, Anthropic, Google, AWS Bedrock, Azure, and 100+ more — swap models without code changes
Reliability primitives: Automatic fallback (configurable by error type), retries with exponential backoff (up to 5x), load balancing across API keys with weights, request timeouts
MCP Gateway (March 2026): Central control plane for MCP servers — single auth layer, per-team access control over which tools are accessible, full observability of every tool call, identity forwarding. Works with Claude Desktop, Cursor, VS Code, and any MCP-compatible client
Observability: OpenTelemetry-based tracing; metrics, dashboards, and a FinOps dashboard for cost attribution per team/workspace
Prompt management: Version-controlled prompt templates with A/B testing
Guardrails: 50+ built-in guardrails for content safety, PII, and cost controls
Virtual key management: Store API keys in Portkey's vault, issue virtual keys to teams, rotate/revoke without changing code
vs. LiteLLM and OpenRouter
Portkey
LiteLLM
OpenRouter
Self-hosted
✅ (open source)
✅ (open source)
❌ (SaaS only)
Observability
Built-in (full OTel)
Basic
Usage stats only
MCP gateway
✅
❌
❌
Best for
Enterprise teams needing governance + observability
