The 3 Best Portkey Alternatives in 2026

LiteLLM is an MIT-licensed SDK and proxy server that puts every major model provider behind one OpenAI-compatible endpoint. Requests go straight from your servers to the providers you hold keys for, so no third party sits in your traffic path. The proxy adds virtual keys with budgets and rate limits per key, user or team, plus fallbacks, load balancing and spend tracking (it needs a Postgres database for those). The open-source edition is free; you pay for the server. LiteLLM Enterprise adds SSO, audit logs, role-based access and support at custom pricing.

• Free and open source (MIT)
• No third party in your request path
• Per-team keys and budgets

Helicone started as LLM request logging and now ships an open-source AI gateway alongside it, with caching, rate limits and automatic fallbacks across providers. Every request is logged with cost, latency and the full prompt, and can be queried with its HQL query language. The Hobby plan is free for 10,000 requests a month with 7-day retention. Pro is $79/month (unlimited seats, alerts, one-month retention) and Team is $799/month (SOC 2, HIPAA, three-month retention), both usage-based after the first 10K requests. You can also self-host it.

• Open source and self-hostable
• Cost and latency logging on every request
• Generous free tier

Cloudflare AI Gateway sits between your app and model providers on Cloudflare's edge. Analytics, caching and rate limiting are free on every plan; you change the base URL and keep your provider keys. Gateways created after Sep 24 2026 store logs under Workers Logs pricing (older gateways keep 100,000 logs on Workers Free and 10 million per gateway on Workers Paid). Guardrails are billed as Workers AI inference. Unified Billing lets you pay providers through Cloudflare for a 5% fee on credits, with no markup on inference.

• Core features are free
• Runs on Cloudflare's global edge
• Nothing to host