Comparison adapter · Official
system-one-adapter-python
Official TypeSafe adapter that speaks the same system_one interface, backed by OpenAI or Anthropic instead of Jev.
- Runnable
- Runnable with Python 3.10+ after installing an extra such as system-one-adapter[openai]. It does not call Jev unless you switch back to TypeSafeClient.
- Keys
- Provider keys (OpenAI or Anthropic), not TYPESAFE_API_KEY, unless you compare against TypeSafeClient yourself.
- Platforms
- Python 3.10+, OpenAI-compatible APIs, Anthropic
- Primitives
- choice, score, noul
pip install 'system-one-adapter[openai]'The adapter lets you keep Noul, Choice, and Score questions in application code while swapping the backend. That is the right way to A/B Jev against a constrained LLM.
Configure llm_answer_mode as probabilities or discrete, and optionally normalize invalid distributions. Provider instances can point at a custom OpenAI-compatible base URL.
Use this when you cannot get a TypeSafe key yet, or when you need a fallback path. Do not advertise LLM adapter latency as Jev latency.
Caveats
- Official TypeSafe repository, but the runtime is an LLM stand-in.
- Useful for cost/speed/intelligence comparisons. It is slower and more expensive than Jev on System One tasks according to TypeSafe's own positioning.
- Requires typesafe-sdk>=0.7.0 for the shared response types.