🇺🇸 USA · LiteLLM

Status: 🟩 COMPLETE 🟦 LIVING Last updated: 2026-06-26 Plain-English tagline: Open-source Python library + self-hosted proxy that lets you call any AI provider (OpenAI, Anthropic, Google, Bedrock, Vertex, Together, 100+ others) using the OpenAI API format. The Swiss Army knife of multi-provider AI.

Front-matter facts

Field	Value
Vendor	BerriAI (USA) — open-source maintained
Country / origin	🇺🇸 USA
Recommended for Australian users?	✅ Yes — open-source, self-hostable, no geographic restrictions
Privacy summary	LiteLLM itself sends nothing — proxies to your chosen provider; self-hosted = full control
Free tier	LiteLLM library + proxy are free (MIT license); pay your AI providers
Paid tiers	None for LiteLLM itself; LiteLLM Enterprise (managed service) quoted
First released	2023
Last reviewed	2026-06-26
Official site	https://litellm.ai

What it is

LiteLLM is an open-source Python library + self-hostable proxy server that gives you a unified interface to 100+ AI providers using the OpenAI API format. Instead of writing different code for OpenAI / Anthropic / Google / Bedrock / Vertex / Together / etc., you write OpenAI-style code; LiteLLM translates.

Two modes:

Python library — import litellm, drop-in for OpenAI SDK
Proxy server — self-host a proxy; configure providers; your apps call the proxy

Use cases:

Multi-provider applications — use Claude / GPT / Gemini interchangeably
Vendor portability — write code that’s not locked to one provider
Cost / spend tracking across providers in one place
Caching to reduce duplicate requests
Rate limiting at the proxy level
Self-hosted observability + control (vs SaaS aggregators)

Differs from OpenRouter (SaaS aggregator):

LiteLLM = self-hosted / library — you control everything
OpenRouter = hosted SaaS — convenience but data flows through OpenRouter

What you’d use it for

Multi-provider apps without SaaS dependency
Enterprise wanting full self-hosted control over AI routing
Cost / spend tracking in one dashboard
A/B testing providers
Fallback / load balancing across providers
Compliance use cases where SaaS aggregators are unacceptable
Self-hosted enterprise AI gateway

How to use

As Python library

pip install litellm
Set provider API keys as env vars (OPENAI_API_KEY, ANTHROPIC_API_KEY, etc.)

Use OpenAI-format calls:

from litellm import completion
response = completion(
    model="anthropic/claude-sonnet-4-6",  # or "openai/gpt-5", "vertex_ai/gemini-3-pro", etc.
    messages=[{"role": "user", "content": "Hello"}]
)

As self-hosted proxy

Deploy LiteLLM proxy (Docker / k8s / VM)
Configure providers in YAML
Your apps call the proxy URL instead of provider URLs
Get unified billing, logging, rate-limit, caching

What it costs

Library + open-source proxy: free (MIT)

LiteLLM Enterprise (managed)

Quoted pricing
SLA, dedicated support, advanced features

Underlying provider APIs

Pay your providers (OpenAI, Anthropic, Google, etc.) at their normal rates
No LiteLLM markup

How it compares to alternatives

Aspect	LiteLLM	OpenRouter	Portkey	Vercel AI SDK	Cloudflare AI Gateway
Open-source	Yes (MIT)	No	No	Yes (Apache 2.0)	No
Self-hosted	Yes	No	Optional	Library	No
Multi-provider	100+	100+	Strong	Strong	Yes
Observability	Self-hosted	Dashboard	Best	Self-instrument	Yes
Caching / rate-limit	Yes	Yes	Yes	Limited	Yes
Cost	Free	Small markup	Tier-based	Free (library)	Free
Best for	Self-hosted full control	Easiest SaaS aggregator	Enterprise observability	TS/React apps	Cloudflare users

For enterprise / compliance / “we control our own AI infrastructure,” LiteLLM is the strongest open-source choice.

Privacy / data handling

LiteLLM library: data goes from your code → provider directly (LiteLLM is just a translator)
LiteLLM proxy: data goes through your self-hosted proxy → provider (you control the proxy)
No data leaves to BerriAI servers in either mode (unless you opt into LiteLLM Enterprise hosted)

Recent changes

2026: Continued provider expansion; ~150 providers supported
2024-25: Major enterprise adoption; observability improvements

Gotchas

Self-hosting overhead for proxy mode — Docker / deployment / monitoring
For simple multi-provider needs, OpenRouter is easier (no infrastructure)
For React / Next.js apps, Vercel AI SDK has cleaner DX
Library mode is excellent for Python apps — start there
Some provider features may not map cleanly to OpenAI format (vendor-specific features)

Tech & AI, Explained

Explorer

litellm