Public Export Index¶
History-specific exports live in kedi.agent_adapter: HistoryProcessorContext,
HistoryProcessorState, HistoryProcessorCallback, HistoryProcessorCondition,
HistoryOrigin, HistoryProcessingError, and
HistoryProcessingValidationError. HistoryProcessingSettings is exported from
kedi.agent_profile. See History.
This index covers the root package's __all__. A root export is not necessarily
a high-level constructor: event unions and protocol aliases describe contracts,
while runtime and artifact handles are advanced integration surfaces.
| Exports | Contract and detailed reference |
|---|---|
__version__ |
Installed Kedi version string; not the provider or grammar revision |
query, bind |
Decorators preserving sync/async call shape; execute docstring/file Kedi, not the Python stub. Query, Bind |
explain, ExplainReport, ExplainedCall |
Read-only per-call-site adapter, model, tool and capability inspection; no user program execution. Run and Parse |
configure, context, reset_config, parallel, interactive |
Configuration replacement, scoped overlay, reset, concurrent override and incremental construction. Parameters |
AdapterLike, AgentName, FrameworkAdapterName |
Selection aliases; framework and harness selectors are not interchangeable. Selection |
type, tool, Constraints |
Class registration, callable metadata/retries/risk, validation-only field metadata. Types and Tools |
Capability |
Reusable tools, guidance, and owned policies applied with > use:; requires supporting adapters. Capability Definitions |
approval, ApprovalPolicy, ApprovalMode, ApprovalHandler, ApprovalRequest, ApprovalDecision, ApprovalCancelled |
Policy registration, static/dynamic policy types, immutable request and allow/deny/edit results; cancellation is not denial. Approvals |
McpServerSpec, SkillsSettings |
Normalized server transport specification and scoped skill-discovery policy. MCP, Skills |
CodeModeSettings |
Activation, preloaded tool names, and limits for the scoped tool catalog. CodeMode |
on, HookSettings, HookHandler, HookEvent, HookEventName, HookDecisionKind |
Registration and event/decision aliases, not arbitrary provider callbacks. Hooks |
HookContext, ToolHookContext, ToolOrigin |
Run lineage and tool-origin metadata supplied by runtime; do not infer authorization from origin. Hooks |
UserPromptSubmitRequest, UserPromptSubmitDecision, PreToolUseRequest, PreToolUseDecision, PostToolUseEvent, PostToolUseFailureEvent, PostToolResultFailureEvent, ToolCallRecord |
Typed hook payloads and body/result evidence; only pre-operation decisions can edit/deny, post events cannot undo effects. Hook payloads |
HookDeniedError, HookExecutionError, HookTimeoutError |
Denial, handler failure, and timeout remain distinct failure paths. Hooks |
AgentMessageEvent, AgentMessagePhase, AgentRunState, AgentRunStateEvent, AgentStreamEvent |
Semantic message/run-state events and discriminants, not token streams. Stream Events |
AsyncAgentEventQueue, observe_agent_events, bind_agent_event_dispatcher |
Async queue and scoped event observer/dispatcher; own and join producer tasks explicitly. Stream Events |
SubagentExecutionContext, SubagentUsageLimits, current_subagent_execution |
Current child lineage (or None outside a child) and request/tool/token ceilings. Python Embedding |
SubagentScope, SubagentTask, SubagentResult |
Invocation-owned child scope, task handle with async wait(), and completed result with typed output. Python Embedding, Native Tasks |
AgentInput, InputReceipt |
Runtime-owned input identity and routing, plus queued, interrupt-requested, or rejected admission receipts; admission does not confirm delivery. Task Input and Redirection |
RunBudget, RunBudgetExceeded |
Immutable root/descendant request and managed tool-attempt ceilings; admission is atomic. Run Budgets |
ConversationState, session |
Model history and artifact ownership, not incremental variable scope. Sessions |
ArtifactPolicy, ArtifactRef, ArtifactChunk, ArtifactSearchResult, ArtifactReleaseResult, ArtifactHandle, ArtifactStream |
Policy, bounded transport/results, internal native handle, explicit single-use stream. Artifacts, Retrieval, Lifetime |
DecisionCapture, DecisionInfo, DecisionSource, capture_decisions, decision_info |
Historical evaluation metadata, capture scope, and binding lookup; no extra model requests. Decisions |
CodexModel, CodexModelAdapter, CodexResponsesConnection, CodexResponsesFallback, codex_responses_model |
Model factory/result aliases, framework choice, HTTP/WebSocket and explicit fallback policy. Codex Models |
CacheInfo, cache_info, clear_cache |
Process-memory cache counts and invalidation; no provider cache control. Cache Control |
KediRuntime, AgentSurface, i, o, c |
Compiled container, bound agent lifecycle, and input/output/call expression constructors. Embedding, Runtime |
KediPromise, KediPromiseLeak, force |
Deferred value, erroneous concrete use and explicit resolution. Promises |
InteractiveSession, dump_session, load_session, SessionPersistenceError, SessionDumpError, SessionLoadError |
Incremental execution and strict snapshot serialization/errors. Incremental Execution |
LoopIterationLimitError |
Conditional-loop guard exceeded, not normal completion. Loops |
Executor, DefaultExecutor, ExecutorDebugEvent, ExecutorDebugExporter, MarkdownDebugExporter, default_debug_export_path |
Python execution protocol and diagnostic exports; default execution is unsandboxed and exports are not secret-redacted. Executors |
kedi.decisions additionally exports Probability, Rubric, BooleanCriteria,
and ChoiceCriteria from the optional kedi-decisions package. Importing these
types requires that package, but ordinary decision-evidence inspection does not.
The criteria are shared by Jev and Laya, without importing either provider SDK;
see Decision Criteria.