llm-adapters

A minimal, auditable Python library for calling multiple LLM providers. Built as a lightweight alternative to LiteLLM after the March 2026 PyPI supply chain attack.

Why?

• Only 2 dependencies: httpx + pydantic — no official SDKs, no bloat
• Each provider is a single file — fully readable and auditable
• All deps pinned with hash verification in CI
• < 2000 lines of code — anyone can review the entire codebase in minutes

Install

pip install llm-dial

Quick Start

from llm_adapters import LLMClient

client = LLMClient()

# Unified interface — auto-routes to the correct provider
response = await client.chat(
    model="openai/gpt-4o",
    messages=[{"role": "user", "content": "Hello!"}],
)
print(response.content)

# Streaming
async for chunk in client.chat_stream(
    model="anthropic/claude-sonnet-4-20250514",
    messages=[{"role": "user", "content": "Tell me a story"}],
):
    print(chunk.content, end="")

Supported Providers

Three Ways to Use

Your app
    │
    ├── As a library: from llm_adapters import LLMClient
    │
    ├── As a proxy: app → localhost:4000 → llm-adapters serve → LLM APIs
    │       (drop-in LiteLLM sidecar replacement, zero code changes)
    │
    └── As a gateway base: app → llm-gateway → llm-adapters → LLM APIs
            (add routing / auth / rate limiting / budget / observability)

Comparison with LiteLLM

Security

This library was born from the LiteLLM supply chain incident. We take dependency security seriously:

• Minimal dependency tree (2 direct deps)
• All versions pinned with SHA-256 hashes in CI
• Automated pip-audit on every PR
• SBOM included in releases
• Published via GitHub Actions OIDC (Trusted Publishers) — no PyPI tokens

License

MIT