lexigram-ai-llm

LLM client layer for the Lexigram Framework — OpenAI, Anthropic, Ollama, Cohere, Groq, Mistral

Overview

LLM client layer for the Lexigram Framework. Provides typed, async-first clients for 18 providers, multi-provider routing, thinking/reasoning control, structured extraction, streaming, embeddings, and model management — all wired through the DI container via LLMModule. Zero-config usage starts with sensible defaults.

Full documentation: docs.lexigram.dev

Install

uv add lexigram-ai-llm
# Optional extras
uv add "lexigram-ai-llm[openai,anthropic,ollama]"

Quick Start

from lexigram import Application
from lexigram.di.module import Module, module

from lexigram.ai.llm import LLMModule
from lexigram.ai.llm.config import ClientConfig

@module(imports=[
    LLMModule.configure(
        ClientConfig(provider="anthropic", model="claude-sonnet-4-6")
    )
])
class AppModule(Module):
    pass

app = Application(modules=[AppModule])
if __name__ == "__main__":
    app.run()

Configuration

Zero-config usage: Call LLMModule.configure() with no arguments to use defaults.

Option 1 — YAML file

# application.yaml
ai_llm:
  provider: "anthropic"
  model: "claude-sonnet-4-6"
  api_key: "${LEX_AI_LLM__API_KEY}"
  temperature: 0.7
  max_tokens: null

Option 2 — Profiles + Environment Variables (recommended)

export LEX_AI_LLM__PROVIDER=anthropic
# Environment variables for each field

Option 3 — Python

from lexigram.ai.llm.config import ClientConfig
from lexigram.ai.llm import LLMModule

config = ClientConfig(
    provider="anthropic",
    model="claude-sonnet-4-6",
)
LLMModule.configure(config)

Config reference

Module Factory Methods

Key Features

• 18 providers: OpenAI, Anthropic, Google Gemini, Azure, Ollama, Groq, Mistral, Cohere, and more
• Multi-provider routing: Sequential, cost-optimized, and latency-optimized strategies
• Thinking/reasoning control: Extended thinking with token budget and suppression
• Structured extraction: JSON schema and Pydantic model extraction
• Streaming: Async streaming response support
• Embeddings: Text embedding client with same provider
• Caching: Response-level caching with configurable TTL

Testing

async with Application.boot(modules=[LLMModule.stub()]) as app:
    # your test code
    ...

Key Source Files