Portkey Ai

Portkey AI is an AI gateway — a single endpoint that routes LLM requests to 100+ providers (OpenAI, Anthropic, Gemini, local models, etc.) with automatic fallbacks, load balancing, response caching, semantic caching, guardrails, and detailed analytics. The gateway is OpenAI API-compatible, meaning a single baseURL change routes existing OpenAI SDK calls through Portkey. For agents needing provider flexibility, cost optimization (cached responses), or reliability (automatic fallback on 429/500): Portkey adds this without code changes. Confidence is docs-derived.

OpenAI-compatible API: change baseURL to api.portkey.ai/v1 and add x-portkey-api-key header. Portkey SDK: from portkey_ai import Portkey; client = Portkey(api_key=..., virtual_key=...). Configs define routing: fallback (primary → backup on failure), load_balance (distribute across providers), retry. Semantic cache: near-duplicate requests served from cache. Guardrails: input/output content policies. Vault: store provider API keys as virtual keys (no key exposure in app code).

Portkey API key + Virtual Keys (provider keys stored in Portkey Vault). API key from Portkey Dashboard. x-portkey-api-key header. Virtual keys reference stored provider credentials without exposing them in application code. HTTPS enforced. Audit logging for all requests.

Gateway-level failure handling: fallback configs automatically retry with alternate provider on 4xx/5xx. Retry configs with exponential backoff. Cache hit rates visible in analytics. Guardrail violations: configurable block or log behavior. Portkey uptime at status.portkey.ai. Requests logged with full payload for debugging.

portkey.ai/docs covers quickstart (baseURL change), virtual key setup, config creation, guardrails, and analytics. Getting started: change one line of code to route through Portkey; immediate observability. Free tier with generous limits. SDK for Python, JS/TS. Community via Portkey Discord.