Quickstart: Strike docs

strike-1 gives a probability on every answer. Strike is in beta, so read Probabilities in beta before you rely on them.

During uncalibrated beta, treat probabilities as directional signals rather than calibrated forecasts. See Probabilities in beta.

Install

bash

pip install strikeflow

Set your API key

Base URL: https://strike-ai.eu (this is the SDK default).

Create an API key in the Strike console (sign in required). Export it:

bash

export STRIKE_API_KEY="your_key_here"

Optional: name a tracing project so decision spans show up grouped in the console.

bash

export STRIKE_PROJECT="my-agent"

First decision

Python

import strikeflow as strike

strike.configure()  # reads STRIKE_API_KEY and STRIKE_API_URL from the environment
result = strike.decide("Should we refund this charge?", "yes_no", context="Customer reports a duplicate charge on invoice 4411.")
print(result.answer, result.confidence, result.trace_id)

result.answer is "yes" or "no". result.probabilities maps every option to a value between 0 and 1 (they sum to 1). result.trace_id identifies the decision span when tracing is enabled.

Next steps

Concepts: choice, score, and yes/no decisions.
decide and adecide: sync and async APIs, decorators, and DecisionResult fields.
Decision tools: expose strike_decision to an LLM agent on OpenAI, Anthropic, Gemini, or MCP.