Module provider 

Modules

claude

ollama

openai

retry

HTTP retry / backoff / Retry-After helpers for LLM providers.

Enums

ReasoningPolicy

Per-provider strategy for echoing back reasoning_content from historical assistant tool_call messages on subsequent requests.

Constants

REASONING_PLACEHOLDER

Sentinel emitted on outbound reasoning_content when we have nothing to echo (cross-provider handoff, pre-fix session, non-thinking model that still tool-called) but the receiving API requires the field to be non-empty (DeepSeek V4 thinking mode rejects empty strings).

Traits

LlmProvider

Functions

create_provider

Factory: create the right provider from config. If api_key is None, automatically loads from $ATOMCODE_HOME/auth.toml (with token refresh if expired).

model_name_suggests_vision

Heuristic: does this model name look like a vision-capable model?

unavailable_provider