Migrating from LiteLLM

llmgate is a lighter, more opinionated alternative to LiteLLM. Here's how to migrate.

API comparison

LiteLLM	llmgate	Notes
litellm.completion(model, messages)	llmgate.completion(model, messages)	Identical call shape
litellm.acompletion(...)	llmgate.acompletion(...)	Identical
response["choices"][0]["message"]["content"]	response.text	Pydantic model, not dict
response["usage"]["total_tokens"]	response.usage.total_tokens	Attribute access
groq/llama-3.1-8b	groq/llama-3.1-8b	Same prefix convention
litellm.exceptions.AuthenticationError	llmgate.exceptions.AuthError	Renamed, slimmer hierarchy
litellm.embedding(model, input)	llmgate.embed(model, input)	Similar shape

---

Find and replace

# Before
import litellm
response = litellm.completion(model, messages)
text = response["choices"][0]["message"]["content"]
tokens = response["usage"]["total_tokens"]

# After
from llmgate import completion
response = completion(model, messages)
text = response.text
tokens = response.usage.total_tokens

---

Error handling

# Before (LiteLLM)
from litellm.exceptions import AuthenticationError, RateLimitError
try:
    ...
except AuthenticationError:
    ...
except RateLimitError:
    ...

# After (llmgate)
from llmgate.exceptions import AuthError, RateLimitError
try:
    ...
except AuthError:
    ...
except RateLimitError:
    ...

---

What llmgate doesn't do

Feature	Status
Gateway / proxy server mode	❌ Out of scope — use a dedicated gateway
LITELLM_LOG environment variable	❌ Use LoggingMiddleware instead
Fallbacks via litellm.completion(fallbacks=[...])	❌ Implement via try/except or a wrapper
Cost tracking	❌ Not yet — planned
100+ providers	❌ 9 providers, all production-quality

llmgate trades breadth for depth — fewer providers, all tested, all typed.