Comparison
LiteLLM is an excellent open-source LLM proxy — if you want to self-host and manage infrastructure. Neureus is the managed alternative: you get the gateway layer plus RAG, agents, workflows, and MCP, deployed on Cloudflare's global edge with zero ops.
LiteLLM is a proxy: it sits in front of your provider keys and adds routing, fallbacks, and virtual keys. That's its whole job — and it does it well. Neureus is an AI application backend: it includes the same routing layer, then adds RAG, ReAct agents, a workflow engine, MCP, HITL, and a marketplace — all managed, all on the edge.
| Feature | N Neureus AI | LiteLLM |
|---|---|---|
| Setup & deployment | ||
| Deployment model | Fully managed — zero infra | Self-hosted (Docker / K8s) |
| Time to first API call | < 2 minutes | 30–60 min (Docker setup) |
| Infra to maintain | None | VPS / K8s cluster |
| Uptime SLA | 99.9% (Cloudflare edge) | You own it |
| Auto-scaling | Built-in (global CDN) | Manual K8s scaling |
| LLM routing | ||
| Providers supported | 9 providers, 35+ models | 100+ providers |
| Load balancing | Yes — automatic | Yes — configurable |
| Fallback routing | Yes — automatic on error | Yes — configurable |
| BYOK (bring your own key) | Yes — AES-GCM encrypted | Yes — env vars / vault |
| OpenAI-compatible API | Yes | Yes |
| Multi-tenancy | ||
| Per-tenant isolation | Yes — every table scoped | Virtual keys (team-level) |
| Per-tenant spend caps | Yes | Yes (budget limits) |
| Per-tenant BYOK keys | Yes | Partial (per virtual key) |
| Org / team grouping | Yes (orgs + RBAC) | Teams in proxy |
| RBAC roles | owner / admin / dev / viewer | admin / user |
| AI capabilities beyond routing | ||
| RAG pipeline | Built-in — ingest, embed, query | ❌ Not included |
| Agent framework | Built-in ReAct loop | ❌ Not included |
| Workflow engine | Yes — Durable Object backed | ❌ Not included |
| Human-in-the-loop (HITL) | Yes — approval steps + DO alarm | ❌ Not included |
| MCP server | Yes — 9 tools, streamable HTTP | ❌ Not included |
| Composable AI patterns | 7 patterns (chain, fan-out, etc) | ❌ Not included |
| Marketplace / templates | 20 templates (5 verticals) | ❌ Not included |
| Vector search | Yes (Vectorize) | ❌ Not included |
| Batch inference (50% off) | Yes — async batch API | ❌ Not included |
| Observability | ||
| Request logging | Analytics Engine — per-request | Langfuse, Helicone integrations |
| Cost tracking | Built-in per-tenant | Via integrations |
| Token usage dashboard | Yes (dashboard) | UI (self-hosted) |
| Alerting | Monitoring room + webhooks | Via Slack / webhook |
| Security & compliance | ||
| PII / PHI guardrails | Built-in (composite profiles) | Via Guardrails AI integration |
| API key auth | Yes — hashed + scoped | Yes — virtual keys |
| WebAuthn / passkeys | Yes | ❌ Not included |
| SSO (OIDC) | Yes — PKCE, RS256/ES256 | Enterprise plan |
| Secrets encryption | AES-GCM per BYOK key | Vault / env-based |
| Pricing | ||
| Free tier | Yes — 5M tokens, no card | Open source (self-host) |
| Managed cloud pricing | From $29/month | $50/month (LiteLLM Cloud) |
| Model cost vs OpenRouter | −10% on every model | Pass-through (0% markup) |
| Server cost (self-hosted) | $0 (managed) | $20–200+/mo (VPS/K8s) |
LiteLLM is free to run (MIT license), but the infrastructure isn't. Add it up before you commit.
Neureus uses an OpenAI-compatible API. Changing your base URL is all it takes to start routing through Neureus instead.
- base_url = "http://localhost:4000" # LiteLLM proxy
- api_key = "sk-your-litellm-virtual-key"
+ base_url = "https://app.neureus.ai" # Neureus
+ api_key = "nk_your-neureus-api-key"
# Everything else stays the same:
response = openai.chat.completions.create(
model = "gpt-4o",
messages = [{"role": "user", "content": "Hello"}]
)LiteLLM supports 100+ providers out of the box — it's a clear strength. Neureus currently supports 9 providers and 35+ models (OpenAI, Anthropic, Google, Meta, Mistral, DeepSeek, Cohere, Qwen, Cloudflare Workers AI). Additional providers are added based on demand. If you need a niche provider, LiteLLM is the better choice today.
Neureus is a managed service running on Cloudflare's global edge. It's not designed for self-hosting. If vendor independence and self-hosting are hard requirements, LiteLLM (open source, MIT license) is the right fit.
Yes. Since Neureus exposes an OpenAI-compatible /ai/chat endpoint, any framework that supports a custom base_url works out of the box — LangChain, LlamaIndex, AutoGen, CrewAI, and others.
LiteLLM virtual keys and Neureus API keys are independent systems. You'll create new API keys in Neureus (nk_ prefix) and update your application config. BYOK provider keys (OpenAI, Anthropic) are re-entered through the Neureus dashboard and stored AES-GCM encrypted.
The Neureus SDK (@neureus/sdk) is open source on npm. The API worker is not open source. The platform runs on Cloudflare Workers — no black box servers, just edge functions. LiteLLM is MIT licensed and fully open source.
Free tier includes 5M AI tokens, RAG pipeline, agents, and all Composable Intelligence patterns. No credit card. No infra.