decide and adecide: Strike docs

decide

strike.decide(question, options, **kwargs) calls strike-1 synchronously.

Python

import strikeflow as strike

result = strike.decide(
    "Which team should handle this?",
    ["billing", "tech", "sales"],
    messages=[{"role": "user", "content": "Refund the duplicate charge on my card."}],
    run_id="ticket-4411",
    host="my-service",
)
print(result.answer, result.probabilities, result.confidence, result.trace_id)

Options argument

`options`	Kind
list of strings	`choice`
`"yes_no"` / `None`	`yes_no`
int (e.g. `5`)	`score` with levels `1`..`n`
`Decision` instance	uses that kind and option list

Force a kind with kind="score" or kind="yes_no" when needed.

Context sources

The model sees a single string state on the wire. The SDK builds it from:

context= string or JSON-serializable dict
messages= chat history (OpenAI dicts, Anthropic blocks, LangChain messages, or plain strings)
state= extra key/value map merged into context

Inside with strike.agent_context(messages=..., state=..., run_id=...):, omitted kwargs are filled automatically.

adecide

Use await strike.adecide(...) in async agents (same parameters as decide).

Python

import asyncio
import strikeflow as strike

async def main():
    result = await strike.adecide(
        "Escalate to a human?",
        "yes_no",
        context="Customer threatens to churn after repeated outages.",
    )
    print(result.answer)

asyncio.run(main())

decides decorator

Build context in your function; the decorator calls strike-1 with that return value.

Python

import strikeflow as strike

@strike.decides("Should we escalate?", "yes_no", min_confidence=0.8)
def escalation_context(ticket):
    return {"ticket_id": ticket["id"], "summary": ticket["summary"]}

result = escalation_context({"id": "t-1", "summary": "Angry customer"}, messages=[...], run_id="run-9")

Passthrough kwargs: messages, state, run_id (when not in the wrapped function signature).

DecisionResult

Field	Meaning
`kind`	`choice`, `score`, or `yes_no`
`answer`	Chosen option label
`probabilities`	Map option → probability (sums to 1)
`confidence`	Probability of `answer`
`prob_source`	`"native"` for strike-1
`score`	Expected score level for `score` decisions
`model`	`strike-1`
`provider`	`strike`
`latency_ms`	Server-reported latency when present
`tokens`	Usage dict when present
`request_id`	API request id
`trace_id`	Decision span id when tracing recorded the call
`run_id`	Host run id when tracing recorded the call
`low_confidence`	True when below `min_confidence`
`yes`	Property: `answer == "yes"`

Serialize for agents:

Python

payload = result.to_dict()

Strike client

For multiple engines or explicit StrikeModel options:

Python

import strikeflow as strike

engine = strike.Strike(model=strike.StrikeModel(model="strike-1", debias=False))
result = engine.decide("Pick one", ["a", "b"], context="demo")

HTTP endpoint used: POST {STRIKE_API_URL}/v1/systemone with Authorization: Bearer <STRIKE_API_KEY>.