Public API index¶

The package root is the stable import surface:

from executionkit import Provider, TokenUsage, consensus

Names listed here come from executionkit.__all__. Import internal modules only when a guide explicitly identifies an advanced extension point.

Patterns¶

Name	Purpose	Reference
`consensus`	Vote across independent completions.	Patterns
`refine_loop`	Generate, score, and revise.	Patterns
`react_loop`	Run a bounded model/tool loop.	Patterns
`structured`	Parse, validate, and repair JSON.	Patterns
`pipe`	Pass one pattern result to the next.	Patterns
`map_reduce`	Map across inputs and reduce the results.	Patterns
`consensus_sync`	Synchronous wrapper for `consensus`.	Patterns
`refine_loop_sync`	Synchronous wrapper for `refine_loop`.	Patterns
`react_loop_sync`	Synchronous wrapper for `react_loop`.	Patterns
`structured_sync`	Synchronous wrapper for `structured`.	Patterns
`pipe_sync`	Synchronous wrapper for `pipe`.	Patterns
`map_reduce_sync`	Synchronous wrapper for `map_reduce`.	Patterns

Name	Purpose	Reference
`PatternResult`	Completed pattern value, score, cost, and metadata.	Patterns
`StreamingPatternResult`	Async text stream with a live cost view.	Patterns
`TokenUsage`	Input-token, output-token, and call counts.	Patterns
`Tool`	Registered async tool definition.	Patterns
`VotingStrategy`	`majority` or `unanimous`.	Patterns
`TerminationReason`	Reason a loop stopped.	Patterns
`Evaluator`	Async response-score callback type.	Patterns
`CheckpointCallback`	Loop checkpoint callback type.	Patterns
`PatternStep`	Callable protocol accepted by `pipe`.	Patterns

Name	Purpose	Reference
`Provider`	OpenAI-compatible chat-completions HTTP client.	Providers
`LLMProvider`	Structural protocol for non-streaming providers.	Providers
`ToolCallingProvider`	Provider protocol with tool-call support.	Providers
`StreamingProvider`	Provider protocol with an async text stream.	Providers
`LLMResponse`	Parsed provider response.	Providers
`ToolCall`	Parsed model-requested tool call.	Providers
`MockProvider`	Scripted provider for tests.	Providers
`ClaudeAgentProvider`	Claude Agent SDK transport authenticated by a Claude subscription sign-in.	Providers

Name	Purpose	Reference
`Kit`	Provider session with cumulative usage and optional conversation state.	Execution helpers
`RetryConfig`	Retry count, delay, exception, and rate-limit policy.	Execution helpers
`DEFAULT_RETRY`	Default retry configuration.	Execution helpers
`TokenBucket`	Async call pacing and cooldown.	Execution helpers
`CostTracker`	Mutable token and call accumulator.	Execution helpers
`estimate_cost`	Apply caller-supplied token rates.	Execution helpers
`ConvergenceDetector`	Score-threshold and stalled-progress detector.	Execution helpers
`extract_json`	Extract a JSON object or array from model text.	Patterns
`validate_score`	Require a finite score in `[0.0, 1.0]`.	Patterns
`checked_complete`	Advanced budgeted/retried completion helper.	Patterns
`checked_stream`	Advanced budgeted/retried streaming helper.	Patterns

Name	Purpose	Reference
`RouteRule`	Named provider-selection predicate.	Execution helpers
`Router`	Select a provider and optionally run a pattern.	Execution helpers
`Step`	Named workflow step with dependencies.	Execution helpers
`Workflow`	Run steps after their dependencies complete.	Execution helpers
`WorkflowCheckpoint`	Workflow progress snapshot with `to_dict()` and `from_dict()`.	Execution helpers
`WorkflowResult`	Workflow outputs and total usage.	Execution helpers
`PlanStep`	Named ordered plan step.	Execution helpers
`Plan`	Run plan steps in order.	Execution helpers
`PlanResult`	Plan outputs and total usage.	Execution helpers
`ApprovalRequest`	Proposed action sent to an approval callback.	Execution helpers
`ApprovalDecision`	Approval result and reason.	Execution helpers
`ApprovalGate`	Sync-or-async approval callback with timeout policy.	Execution helpers

Name	Purpose	Reference
`TraceEvent`	Event kind and read-only payload.	Execution helpers
`TraceCallback`	Sync-or-async event callback type.	Execution helpers
`emit_trace`	Invoke a trace callback when present.	Execution helpers
`CallRecord`	One logical LLM call reconstructed from correlated trace events.	Execution helpers
`TraceCollector`	Trace callback building per-call records and portable run summaries.	Execution helpers

Name	Purpose	Reference
`EvalCase`	One run/check pair.	Evaluation
`EvalResult`	One case result.	Evaluation
`EvalReport`	Aggregate results and accuracy.	Evaluation
`run_eval_suite`	Run cases in order.	Evaluation
`Turn`	One checked conversation turn.	Evaluation
`ConversationScript`	Ordered conversation turns.	Evaluation
`run_conversation_script`	Evaluate a script with one `Kit`.	Evaluation
`live_provider_from_env`	Build an opt-in live provider from environment variables.	Evaluation

Name	Purpose	Reference
`AnthropicBatchClient`	Minimal native Message Batches HTTP client.	Batch API
`consensus_batch`	Run consensus samples as one batch job.	Batch API
`map_batch`	Submit several prompts and return ordered responses.	Batch API

Name	Meaning	Reference
`ExecutionKitError`	Base package exception.	Execution helpers
`LLMError`	Base provider-call exception.	Execution helpers
`RateLimitError`	HTTP 429 with `retry_after`.	Execution helpers
`PermanentError`	Non-retryable provider error.	Execution helpers
`ProviderError`	Retryable transport or provider error.	Execution helpers
`PatternError`	Base pattern exception.	Execution helpers
`BudgetExhaustedError`	A token or call budget blocked another attempt.	Execution helpers
`ConsensusFailedError`	Unanimous consensus was not reached.	Execution helpers
`MaxIterationsError`	A bounded loop ended without a final answer.	Execution helpers
`ApprovalDeniedError`	A required approval was denied.	Execution helpers
`ApprovalTimeoutError`	An approval timed out under the `raise` policy.	Execution helpers

__version__ is the installed package version.