Concepts: Strike docs

Ask strike-1 a question with fixed options and get back one option and a probability on every answer. Your own LLM still handles open-ended text; whenever the workflow needs a choice, a score, or yes/no, call strike-1.

Beta

Strike and strike-1 are in beta. strike-1 gives a probability on every answer. In beta, treat those numbers as routing and debugging signals.

Core objects

Idea	Where it lives
Decision kinds	`choice`, `score`, `yes_no` via `strike.decide(...)` options or `strike.Choice` / `strike.Score` / `strike.YesNo`
Result	`strike.DecisionResult`: `answer`, `probabilities`, `confidence`, `trace_id`, and more
Agent tool	`strike.strike_decision_tool()`: same decision, shaped for LLM tool APIs
Traces	`strike.trace(...)`: one run graph exported to `POST /v1/traces`

Mental model

text

Host agent run
  ├─ LLM / tools / your code
  └─ strike-1 decision  →  answer + probabilities + trace_id

Deep dives:

Typed decisions
Probabilities and confidence
Probabilities in beta
Danger zone: preview for images, a box, and a longer window