Agent • AI Agent SDK
Agent: Class reference for Agent
Agent
Defined in the agent module.
AI Agent
Constructor
Optional[str]No description available.
Optional[str]No description available.
Optional[str]No description available.
Optional[str]No description available.
Optional[str]No description available.
OptionalNo description available.
OptionalNo description available.
Optional[str]No description available.
Optional[str]No description available.
Optional[str]No description available.
OptionalNo description available.
OptionalNo description available.
Methods
Create an Agent from a template.
stream_emitter()Lazy-loaded StreamEventEmitter for real-time events (zero overhead when not used).
stream_emitter()Allow setting stream_emitter directly.
llm_instance()Instance method.
llm_instance()Instance method.
clone_for_channel()Return a fully independent copy of this agent for a gateway channel.
chat_history()Get chat history (read-only access, use context managers for modifications).
chat_history()Set chat history (updates the underlying async-safe state with lock).
auto_memory()AutoMemory instance for automatic memory extraction.
auto_memory()Instance method.
policy()PolicyEngine instance for execution control.
policy()Instance method.
output_style()OutputStyle instance for response formatting.
output_style()Instance method.
thinking_budget()ThinkingBudget instance for extended thinking control.
thinking_budget()Instance method.
reasoning_effort()Unified, provider-portable reasoning-effort level (Issue #4452).
reasoning_effort()Instance method.
total_cost()Cumulative USD cost of all LLM calls in this agent run.
cost_summary()Summary of cost and token usage.
context_manager()ContextManager instance for unified context management.
context_manager()Set context manager directly.
console()Lazily initialize Rich Console only when needed AND verbose is True.
skill_manager()Lazily initialize SkillManager only when skills are accessed.
get_skills_prompt()Get the XML prompt for available skills.
last_stop_reason()Structured reason the last tool-execution loop stopped.
agent_id()Lazily generate agent ID when first accessed.
display_name()Safe display name that never returns None.
set_snapshot_root()Root filesystem change-tracking at `project_path`.
Undo the last set of file changes.
redo()Redo a previously undone set of file changes.
diff()Get file diffs from autonomous execution.
analyze_prompt()Analyze prompt for autonomy signals.
get_recommended_stage()Get recommended execution stage for prompt.
run_autonomous()Run an autonomous task execution loop.
run_autonomous_async()Async variant of run_autonomous() for concurrent agent execution.
handoff_to()Programmatically hand off a task to another agent.
run_until()Run agent iteratively until output meets quality criteria.
run_until_async()Async version of run_until().
handoff_to_async()Asynchronously hand off a task to another agent.
add_verification()Register a verification hook for the current run.
remove_verification()Remove a verification hook by name. Returns True if one was removed.
get_available_tools()Get tools available to this agent, filtered by plan_mode if enabled.
add_mcp_server()Attach an MCP server to a live agent at runtime.
remove_mcp_server()Disconnect and deregister a previously attached MCP server.
refresh_tools()Re-derive the live toolset from `self.tools`.
Return the names of MCP servers currently attached at runtime.
rules_manager()Lazy-initialized RulesManager for persistent rules/instructions.
get_rules_context()Get rules context for the current conversation.
get_memory_context()Get memory context for the current conversation.
get_learn_context()Get learning context for injection into system prompt.
store_memory()Store content in memory.
llm_model()Unified property to get the LLM model regardless of configuration type.
retrieval_config()Get the unified retrieval configuration.
rag()Lazy-loaded RAG instance for advanced retrieval with citations.
retrieve()Retrieve context from knowledge without LLM generation.
query()Query knowledge and get a structured answer with citations.
rag_query()Query knowledge using RAG pipeline with citations.
chat_with_context()Chat with pre-retrieved context.
guardrail_retry_count()How many times output validation has sent the model back for a retry.
last_guardrail_error()The most recent rejection message a guardrail sent back to the model.
close()Synchronously close the agent and clean up resources.
aclose()Async version of close() for async context managers.
is_closed()Returns True if the agent has been closed.
where_does_it_run()Plain-English answer to ‘where does my code actually run?’.
Source
View on GitHubpraisonaiagents/agent/agent.py at line 250