Skip to main content

arp-llm

arp-llm is a shared helper for calling LLM providers in a consistent way across the ARP/JARVIS stack.

It provides two primary interfaces:

  • ChatModel.response(...) (text/chat + optional JSON Schema structured output)
  • Embedder.embed(...) (embeddings)

Where it’s used in JARVIS (v0.3.x)​

  • Selection Service: rank NodeTypes into bounded candidate sets
  • Composite Executor: planning (decomposition) and argument generation

Configuration model​

arp-llm uses an env-driven “profile” pattern:

  • default profile: openai (requires ARP_LLM_API_KEY + ARP_LLM_CHAT_MODEL)
  • optional profile: dev-mock (no network; requires explicit ARP_LLM_PROFILE=dev-mock)

See:

  • GitHub: AgentRuntimeProtocol/ARP_LLM
  • PyPI: arp-llm