Decision#

class scikitplot.agent.Decision(kind, action=None, answer=None, rationale='', usage=None)[source]#

What the planner decided: call a tool, or finish with an answer.

Build one with act or finish.

Parameters:

kind{“act”, “finish”}: Call a tool, or finish with an answer.
actionAction or None, default=None: The call, for an act decision.
answerobject, default=None: The answer, a JSON value, for a finish decision.
rationalestr, default=””: Why, in the planner’s words; recorded, never acted on.
usageUsage or None, default=None: What the decision cost. Required when the run has a token limit.

Parameters:

kind (str)
action (Action | None)
answer (Any)
rationale (str)
usage (Usage | None)

Examples

>>> from scikitplot.agent import Decision
>>> Decision.act("search", {"query": "spans"}).kind
'act'
>>> Decision.finish("42", rationale="found it").kind
'finish'

classmethod act(tool, arguments=None, *, rationale='', usage=None)[source]#

Decide to call tool with arguments.

Parameters:

tool (str)
arguments (Mapping[str, Any] | None)
rationale (str)
usage (Usage | None)

Return type:

Decision

action: Action | None = None#

answer: Any = None#

classmethod finish(answer, *, rationale='', usage=None)[source]#

Decide to finish with answer (a JSON value).

Parameters:

answer (Any)
rationale (str)
usage (Usage | None)

Return type:

Decision

classmethod from_dict(data)[source]#

Rebuild a decision written by to_dict.

Parameters:: data (Mapping[str, Any])
Return type:: Decision

kind: str[source]#

rationale: str = ''#

to_dict()[source]#

Return a JSON-compatible mapping.

Return type:: dict[str, Any]

usage: Usage | None = None#