Agent Loop docs
agentloop
The package root collects common imports. Agent runs conversations, tool defines local
Python tools, Transcript holds messages, and ConversationSummary reports how a run ended.
Import these names directly from agentloop:
from agentloop import Agent, ConversationSummary, Transcript, tool
It re-exports:
- Agent layer:
Agent,Deps,tool. - Configuration:
Settings,build_client. - Errors:
AgentLoopError,ToolError. - Loop:
run_conversation,run_conversation_result,ListObserver. - Observers:
LoopObserver,NullObserver,ConsoleObserver,MultiObserver. - Reporting:
AttemptRecord,ConversationContext,ConversationResult,ConversationStatus,ConversationSummary,FailureCount,SummaryCollector,SummaryOptions,TurnRecord, andEventTraceObserver. - Hosted tools:
HostedTool,HostedToolPart,Source,Citation. - Tools:
Registry,ToolDeps,ToolSpec,allow_all. - Transcripts:
Transcript,Origin,final_text,join_text.
__version__ is the package's declared version string.
This list is an import map; the API index links to method and field details.
Custom providers use LLMClient, ModelRequest, and ModelResponse from
agentloop.provider. Provider-specific clients, the full error hierarchy,
retry configuration, OpenTelemetry setup, and metrics remain in their defining modules.
Importing agentloop does not load optional provider SDKs or the OpenTelemetry SDK.