ARP LLM (arp-llm)

Shared helper for calling LLM providers across the ARP/JARVIS stack.

This package provides:

• ChatModel.response(...) (chat/text + optional JSON Schema structured output)
• Embedder.embed(...) (embeddings)
• load_chat_model_from_env(...) / load_embedder_from_env(...) profile-based configuration

Install

pip install arp-llm

Quickstart (OpenAI default)

export ARP_LLM_API_KEY=...
export ARP_LLM_CHAT_MODEL=gpt-4.1-mini
# Optional (OpenAI is the default profile):
# export ARP_LLM_PROFILE=openai

import asyncio

from arp_llm import Message, load_chat_model_from_env

async def main() -> None:
    model = load_chat_model_from_env()
    resp = await model.response([Message.user("hello")])
    print(resp.text)

asyncio.run(main())

Dev mock (optional; no network)

export ARP_LLM_PROFILE=dev-mock
# Optional: provide deterministic fixtures
# export ARP_LLM_DEV_MOCK_FIXTURES_PATH=./fixtures.json

Configuration (OpenAI)

# ARP_LLM_PROFILE=openai is optional (default)
export ARP_LLM_API_KEY=...
export ARP_LLM_CHAT_MODEL=gpt-4.1-mini
# Optional overrides:
export ARP_LLM_BASE_URL=https://api.openai.com

API

• ChatModel.response(messages, *, response_schema=None, temperature=None, timeout_seconds=None, metadata=None) -> Response
  • If response_schema is provided, Response.parsed will be a JSON-like dict.
• Embedder.embed(texts, *, timeout_seconds=None, metadata=None) -> EmbeddingResponse

Direct construction (advanced)

The load_*_from_env*() helpers are optional. For multi-provider routing/fallback inside a single process, construct provider clients directly (and route per call), for example:

from arp_llm.providers.openai import OpenAIChatModel

model = OpenAIChatModel(model="gpt-4.1-mini", api_key="...", base_url="https://api.openai.com")

See https://github.com/AgentRuntimeProtocol/BusinessDocs/blob/main/Business_Docs/JARVIS/LLMProvider/HLD.md and https://github.com/AgentRuntimeProtocol/BusinessDocs/blob/main/Business_Docs/JARVIS/LLMProvider/LLD.md for the design intent.