fairlane.systems

LITELLM · OPENROUTER · PORTKEY · DUEL

LiteLLM vs OpenRouter vs Portkey – which LLM gateway?

Three LLM gateways head-to-head: self-host vs cloud, number of providers, compliance routing, EU hosting. Decision guide for multi-LLM setups in Swiss SMEs.

Researched & fact-checked by: · As of: 2026-05

What this is about

LiteLLM, OpenRouter, and Portkey are the three most important LLM gateways in May 2026. An LLM gateway sits between your application and the LLM providers (OpenAI, Anthropic, Mistral, Cohere, Google, local Ollama). A single API schema covers all providers – you switch models without code changes.

LiteLLM is a Python open-source project (MIT licence). Self-host common (Docker container), supports 100+ providers, fully OpenAI-compatible. Cloud tier optional. At fairlane.systems we have run LiteLLM in production since 2025 for Swiss fiduciary clients.

OpenRouter is a US SaaS provider – one API key for over 200 models including all proprietary and open-source variants. Very easy onboarding, transparent pay-per-use pricing. Cloud-only, US-hosting (no dedicated EU tier as of May 2026).

Portkey is an enterprise gateway from the US, with cloud service and self-host tier. Offers built-in observability, semantic caching, guardrails, cost tracking, and multi-tenant administration. EU region available; pricing subscription-based.

As of May 2026 all three are production-ready, but target audiences and compliance profiles differ clearly.

Why this comparison?

Without a gateway, multi-LLM routing in a Swiss SME application is painful: every provider has its own SDK, own auth scheme, own quirks. A provider switch = code switch. With a gateway: one schema, all providers.

The gateway choice drives four hard consequences.

First data residency. LiteLLM self-host runs on your EU/CH infrastructure – no third-country hop. OpenRouter routes through US servers, every request passes through (even if the model is at Mistral Paris). Portkey has an EU region (Frankfurt) since Q3 2025; without the EU plan traffic flows through the US main region.

Second compliance routing. For Swiss fiduciary or law firms a rule "data under professional secrecy = EU-region models only, otherwise OpenAI/Anthropic allowed" must be enforceable inside the gateway. LiteLLM and Portkey support routing rules. OpenRouter offers this via a "Preferred Provider" setting, but less granularly.

Third observability. Anyone making LLM calls for 50+ clients needs per-client cost, audit trail, anomaly detection. Portkey has this built in. LiteLLM delivers base logging plus Loki/Grafana integration (you set it up). OpenRouter shows aggregate usage only.

Fourth pricing model. LiteLLM costs server hours (Hetzner CX22 about EUR 4/month). OpenRouter adds about 5 percent markup on vendor prices plus its own cloud margin. Portkey has a subscription from USD 49/month plus provider costs.

Head-to-head on 6 axes

Data protection and EU hosting. LiteLLM wins: full self-host, data stays in your data centre (Hetzner, Exoscale, Infomaniak). Portkey EU region available (Frankfurt) on enterprise tier; standard tier runs US-central. OpenRouter as of May 2026 has no dedicated EU tier, all requests flow through US servers.

Price. LiteLLM server: EUR 4-20/month for a Hetzner container, depending on load. OpenRouter: pay-per-use, vendor price plus about 5 percent markup (e.g. GPT-4o USD 2.50 in / 10 out becomes about USD 2.65 / 10.50). Portkey: subscription from USD 49/month (Starter, 10000 requests), USD 199/month (Production, 100000 requests), Enterprise by quote – plus provider costs.

Number of providers / models. OpenRouter leads: 200+ models including proprietary and open source, lobbies new models in early. LiteLLM: 100+ providers incl. local Ollama, OpenRouter itself, Bedrock, Azure, Vertex. Portkey: 100+ models over 30+ providers, plus its own caching layer.

Compliance routing. Portkey leads: visual routing rules in the UI, guardrails (PII filter, prompt-injection protection), audit log for compliance evidence. LiteLLM: routing rules in Python code or YAML, fully configurable but no UI. OpenRouter: simple preferred-provider setting, no guardrail layer.

Lock-in. LiteLLM: no lock-in, MIT licence, code open, data in your DB. Portkey: lock-in to the Portkey platform; migration to LiteLLM possible (same OpenAI schema), but configuration must be ported by hand. OpenRouter: low technical lock-in (OpenAI schema), but economic markup mandatory.

Operations effort. OpenRouter is easiest: get an API key, done. Portkey: account, subscription, config – about 2 hours. LiteLLM: Docker container, Postgres backend, virtual keys, Loki – about 4-8 hours setup, then 2 hours/month operations.

Decision path in 6 steps

  1. 01Data classification: is the data under professional secrecy or strict Swiss DPA? If yes, LiteLLM self-host is mandatory candidate.
  2. 02Count providers: one = no gateway, 2-5 = LiteLLM or OpenRouter, 5+ with compliance reports = Portkey.
  3. 03Estimate volume: requests per month. Under 10k = OpenRouter (zero ops). Over 10k = LiteLLM or Portkey.
  4. 04Check compliance reports: audit trail, per-client cost, PII filter needed? If yes, Portkey or LiteLLM with Loki/Grafana.
  5. 05Check setup know-how: Docker + Postgres + routing YAML possible? If yes, LiteLLM. If no, OpenRouter or Portkey.
  6. 06Build a pilot: 14 days production, measure 100+ real queries, document latency and cost, then scale.

When which gateway

When LiteLLM. You are a Swiss fiduciary, lawyer, insurer, or doctor. Your data must not see any third country (not even routing servers). You want full control over routing rules, logging, audit trail. You have a sysadmin or provider for Docker. You want no monthly subscription markup on vendor prices. LiteLLM is the default for regulated Swiss SMEs with DPA and professional-secrecy obligations.

When OpenRouter. You are a solo developer, early-stage startup, or marketing agency with standard data. You want to get started in 5 minutes without running a server. You experiment with 10+ models per week and value "one key, 200 models" convenience. You are price-sensitive on the vendor listing but not the markup. You have no data residency requirement. OpenRouter is the default for prototypes, hobby projects, English-language use cases.

When Portkey. You are a larger company with 100+ employees and 10+ AI use cases in parallel. You need built-in observability, per-client cost, audit reports for compliance. You want a vendor with enterprise support and SOC 2 or ISO 27001. Your budget has 200-2000 USD/month for the gateway itself. You have no engineering team willing to keep self-host LiteLLM operationally stable. Portkey is the choice for mid-sized companies with enterprise demands.

When NONE. If you use a single LLM provider (e.g. only OpenAI), no gateway needed – direct SDK is enough. If your load is very small (under 1000 requests per month) and you have no routing complexity, any gateway is over-engineering. If you need absolute tier-A sovereignty and use only local models, Ollama or vLLM directly is the better endpoint.

When NONE of the three

If you use a single provider and that will not change (e.g. a pure OpenAI app with function calling), the gateway layer only adds complexity without value. Direct OpenAI SDK is simpler.

If your LLM load is very low (under 100 requests per day) and you have no compliance requirement, any gateway is over-engineering. Build a simple Python module with two if-branches – while the load allows it.

If you use exclusively local models (Ollama, vLLM), LiteLLM is possible but a direct Ollama API call is often enough. Gateway value emerges only when multiple providers are combined.

If you have a very exotic requirement (e.g. realtime voice streaming via WebSocket, bidirectional RPC with OpenAI), all three gateways support that only partially. Use direct SDK calls at the affected endpoint, otherwise latency and streaming capability are lost.

If your compliance profile is so strict that even the gateway server cannot be in your hand (FINMA tier A, federal infrastructure), you need a hardware-gated setup – even LiteLLM self-host must then run in an air-gapped zone, combined with dedicated audit pipelines.

Trade-offs

STRENGTHS

  • Three-gateway picture covers self-host to enterprise spectrum
  • Compliance routing axis as main criterion for Swiss SMEs explicit
  • Pricing made comparable in EUR and USD instead of vendor marketing talk
  • Clear "pilot in 14 days" sequence instead of vague recommendation

WEAKNESSES

  • Gateway market moves fast – Portkey tiers and OpenRouter markup may change quarterly
  • LiteLLM operations effort is often underestimated (Postgres, backup, updates)
  • Cloud-only gateways (OpenRouter) are barely workable for regulated mandates in May 2026
  • Own Loki/Grafana pipeline for per-client cost is not trivial

FAQ

Can I switch from OpenRouter to LiteLLM?

Yes, no code change. Both use OpenAI-compatible schema. Just swap API base URL and auth key. LiteLLM is OpenRouter-compatible and can treat OpenRouter as one of its providers if you want to migrate step by step.

Is the Portkey subscription worth it for Swiss SMEs?

Only with a clear enterprise requirement. For a 5-person fiduciary office, LiteLLM self-host plus Grafana is more economic (EUR 10-20/month instead of USD 199). Portkey becomes attractive from about 50 employees and 10+ use cases when compliance reports and multi-tenant administration would otherwise be manual.

Does OpenRouter really only add 5 percent markup?

As of May 2026 OpenRouter publishes a small BYOK fee per token – typical markup across all models is in the 4-6 percent range. With direct enterprise contracts (e.g. Anthropic enterprise tier) OpenRouter is more expensive; on standard list prices, negligible.

Where do I run LiteLLM self-host in Switzerland?

Hetzner Falkenstein (EUR 4-20/month) covers EU residency; Exoscale Zurich (CHF 20+/month) and Infomaniak Geneva (CHF 15+/month) deliver pure CH residency. LiteLLM container plus Postgres is enough; Loki/Grafana optional for audit trail.

Related topics

LITELLM · TECHLiteLLM: one gateway for 100+ LLM providers behind a single APIMULTI-LLM GATEWAY · SERVICEMulti-LLM Gateway: eight providers, one entry point, compliance routingROUTING · AI CONCEPTMulti-LLM routing: which model when, for how muchOPENAI · LLM PROVIDEROpenAI GPT models from a Swiss fiduciary perspective: residency, pricing, complianceANTHROPIC · LLM PROVIDERAnthropic Claude from a Swiss fiduciary perspective: residency, pricing, complianceMISTRAL · LLM PROVIDERMistral AI from a Swiss fiduciary perspective: EU residency, pricing, sovereigntyHETZNER · TECHHetzner as EU hosting for Swiss fiduciaries and SMEs: data centres, contracts, cost

Sources

  1. LiteLLM – GitHub repository · 2026-05
  2. OpenRouter – Pricing and providers · 2026-05
  3. Portkey – Documentation · 2026-05
  4. ToolHalla – OpenRouter vs LiteLLM vs Portkey 2026 · 2026-04
  5. PkgPulse – Portkey vs LiteLLM vs OpenRouter LLM Gateway 2026 · 2026-03

FITS YOUR STACK?

What this looks like in your business – a 30-minute intro call.

Book a call