Portkey vs LiteLLM: open-source gateway with a hosted dashboard, or free proxy you run end to end?

Choose LiteLLM when you want the whole gateway, including virtual keys, team budgets, rate limits, fallbacks, request logging and Prometheus metrics, for $0 and are happy to run PostgreSQL (plus optional Redis) yourself. Its Enterprise edition is priced annually by request capacity, never per token, and adds SSO/SCIM, audit logs, secret managers and SLAs.

Choose Portkey when observability is the point: you want logs, traces, feedback, alerts, prompt templates with versioning and guardrails in a hosted dashboard without building that layer. The Developer plan is free with 10k recorded logs a month but is not meant for production; Production is $49 a month for 100k logs, with $9 per extra 100k.

LiteLLM: the open-source proxy is free forever when self-hosted. There is no per-token platform fee. Enterprise pricing is annual and based on gateway request capacity; a 30-day trial key is available without a credit card.

Portkey: the open-source gateway is free to self-host. The hosted platform is tiered by recorded logs: Developer at $0 (10k logs a month, 3-day retention), Production at $49 a month (100k logs, 30-day retention, guardrails, RBAC, semantic caching) and Enterprise on custom terms with SSO, VPC hosting and SOC 2 or HIPAA support. Log volume, not token volume, drives the bill.

LiteLLM covers 100+ providers behind one OpenAI-compatible API and bundles governance into the proxy: virtual keys, users and teams, spend tracking and budgets, rate limits, model fallbacks, request and response logging, and Prometheus metrics. The trade-off is that the proxy needs a PostgreSQL database and, for high throughput, Redis.

Portkey's gateway handles fallbacks, load balancing, retries and timeouts through a universal OpenAI-compatible API. Its strength is the hosted layer on top: logs and traces with metadata filters, feedback, alerts, prompt templates with versioning, simple and semantic caching, and guardrails. Virtual keys, role-based access control and budgets are on the paid plans.

With LiteLLM you own the full stack: upgrades, database backups, uptime and security are yours. Pin package versions and verify checksums: on March 24, 2026 two PyPI releases of the litellm package (1.82.7 and 1.82.8) were published with credential-stealing malware for about 40 minutes, and BerriAI's advisory lists the steps to check an install.

With Portkey you can self-host the open-source gateway, but the dashboard, logs and prompt management stay in Portkey's hosted service unless you are on an Enterprise VPC deployment. Portkey's own guidance says the free Developer plan is for prototyping, not production traffic.

Platform teams putting paid provider keys behind one endpoint, who need per-team budgets and can run infrastructure, get the most from LiteLLM. Its free tier is complete rather than a trial.

Product teams shipping LLM features who would otherwise build their own logging, prompt versioning and guardrail tooling save the most time with Portkey, as long as the monthly log volume fits the $49 Production plan or an Enterprise agreement.

If neither team wants to run anything, a hosted gateway such as OpenRouter or Requesty removes the server entirely in exchange for a platform fee; see the OpenRouter vs LiteLLM and Requesty vs OpenRouter comparisons.