Praison AI

praisonaiagents SDK

Documentation for the praisonaiagents package – Core agent framework

Module praisonaiagents

The praisonaiagents package provides a lightweight, high-performance framework for building AI agents with support for multi-agent orchestration, memory, knowledge bases, and tool integration.

Quick Start

python
from praisonaiagents import Agent

agent = Agent(instructions="You are a helpful assistant")
response = agent.start("Hello!")
print(response)

Core Modules

Agent

Core agent class for AI interactions

 Agents

Multi-agent orchestration

 Task

Task definition and management

 Process

Task execution flows

Agent Variants

ImageAgent

Vision and image processing

 RouterAgent

Intelligent request routing

 DeepResearchAgent

Multi-step research with citations

 QueryRewriterAgent

Query optimization for RAG

 PromptExpanderAgent

Prompt enhancement

 ContextAgent

Context-aware processing

Data & Memory

Memory

Memory management for stateful agents

 Knowledge

Knowledge base and RAG support

 Session

Session management and persistence

 DB

Database adapters and storage

 Checkpoints

State checkpointing

 Context

Context management

Tools & Integration

Tools

Tool system and decorators

 MCP

Model Context Protocol support

 Handoff

Agent-to-agent handoffs

 Workflows

Multi-step workflow execution

 Skills

Agent Skills standard support

 LLM

LLM configuration and providers

Safety & Control

Guardrails

Input/output validation

 Planning

Planning mode support

 Policy

Policy-based controls

 Approval

Human-in-the-loop approval

 Thinking

Extended thinking support

 Compaction

Context compaction

Observability

Telemetry

Observability and monitoring

 Hooks

Event hooks and callbacks

 Display

Console display utilities

 Eval

Agent evaluation framework

 Outputs

Data classes for task outputs

 FlowDisplay

Workflow visualization

UI & Output

UI

UI components (AGUI, A2A)

 Output

Output formatting

 LSP

Language Server Protocol

 Background

Background task execution

Import Patterns

The SDK supports multiple import styles for different use cases:

Simple Import (Recommended for Quick Start)

python
from praisonaiagents import Agent, AgentTeam, Task, tool, Tools

Organized Imports (Recommended for Projects)

For larger projects, use sub-package imports for cleaner code organization:

python
# Core classes from root
from praisonaiagents import Agent, AgentTeam, Task

# Configuration from config sub-package
from praisonaiagents import MemoryConfig, OutputConfig, ExecutionConfig

# Tools from tools sub-package
from praisonaiagents import tool, BaseTool, Tools

# Memory from memory sub-package
from praisonaiagents import Memory, FileMemory

# Workflows from workflows sub-package
from praisonaiagents import AgentFlow, Pipeline, Route, Parallel

Namespace Style

Use the pa. namespace for exploratory development:

python
import praisonaiagents as pa

agent = pa.Agent(instructions="...")
config = pa.config.MemoryConfig(user_id="u123")
memory = pa.memory.Memory(config=config)

Backwards Compatibility

All legacy imports continue to work via lazy loading:

python
# These all work (backwards compatible)
from praisonaiagents import MemoryConfig  # via __getattr__
from praisonaiagents import AgentFlow      # via __getattr__
from praisonaiagents import Memory        # via __getattr__

Sub-Packages Reference

Sub-PackageKey ExportsDescription
praisonaiagents.configMemoryConfig, OutputConfig, ExecutionConfig, PlanningConfig, etc.Feature configuration classes
praisonaiagents.toolstool, BaseTool, Tools, ToolResult, FunctionToolTool creation and management
praisonaiagents.memoryMemory, FileMemory, RulesManagerMemory and persistence
praisonaiagents.workflowsWorkflow, Pipeline, Route, Parallel, LoopWorkflow orchestration

Installation

bash
# Core package
pip install praisonaiagents

# With memory support
pip install "praisonaiagents[memory]"

# With knowledge/RAG support  
pip install "praisonaiagents[knowledge]"

# With MCP support
pip install "praisonaiagents[mcp]"

See Also