llmsays 0.1.1

One-line LLM queries with smart routing and provider failover

llmsays

One-line LLM calls with automatic prompt-tier routing and provider failover.

llmsays keeps usage simple:

from llmsays import llmsays

response = llmsays("Explain quantum tunneling in simple words")
print(response)

Why llmsays

• Single function API: llmsays(prompt)
• Smart routing with sentence-transformers/paraphrase-MiniLM-L3-v2
• Tier selection: small, medium, large, extra_large
• Provider failover: Groq, NIM, OpenRouter, Fireworks, Baseten
• Latency-aware provider ordering
• Optional parallel provider querying for faster first-response

Installation

pip install llmsays

Required Environment Variables

Set at least one provider key (multiple keys recommended for failover):

• GROQ_API_KEY
• OPENROUTER_API_KEY
• NIVIDIA_API_KEY
• FIREWORKSAI_API_KEY
• BASETEN_API_KEY

Example:

export GROQ_API_KEY="your_key"
export OPENROUTER_API_KEY="your_key"

Quick Start

from llmsays import llmsays

user_prompt = input("Here goes your prompt: ")
print(llmsays(user_prompt))

Advanced Usage

Choose provider order:

from llmsays import llmsays

print(
	llmsays(
		"Analyze this legal clause",
		provider_preference=["Groq", "Openrouter", "Fireworks"],
	)
)

Enable parallel provider queries (returns first successful response):

from llmsays import llmsays

print(
	llmsays(
		"Design a production-ready architecture with tradeoffs",
		use_multiprocessing=True,
	)
)

CLI Usage

llmsays "Explain transformers in simple terms"
llmsays "Analyze this legal clause" --providers Groq Openrouter
llmsays "Summarize this API contract" --use-multiprocessing

How Routing Works

1. Heuristic pre-filter estimates complexity quickly.
2. Semantic routing refines tier selection.
3. Selected tier maps to provider-specific model choices.
4. If one provider fails, the next provider is attempted automatically.

Notes

• Requires Python >=3.9
• Internet connection is required to call provider APIs
• Responses depend on the configured provider/model availability

License

MIT