Package Tier Architecture

How praisonaiagents, praisonai-code, praisonai-bot, praisonai-train, praisonai-browser, praisonai-mcp, and praisonai layer on top of each other

PraisonAI ships as seven packages that stack — pick the smallest one that covers what you need.

Quick Start

1. SDK only — embed agents in your Python app

   pip install praisonaiagents
   

   The core Agent, tools, memory, and hooks. No CLI, no gateway.

2. Terminal CLI — run/chat/code, no gateway or bots

   pip install praisonai-code
   

   The agentic CLI (run, chat, code) and LLM runtime, without channel bots.

3. Gateway + bots — channel runtime without the wrapper

   pip install "praisonai-bot[gateway,bot]"
   

   The WebSocket gateway, Telegram/Discord/Slack bots, and the scheduler tick.

4. Training — fine-tune LLMs or train agents

   pip install "praisonai-train[llm]"
   

   LLM fine-tuning via Unsloth and iterative agent training, without the wrapper.

5. Browser — automate Chrome and the web

   pip install "praisonai-browser[all]"
   

   The bridge server, CDP agent, and Playwright engine — usable on their own or inside the full praisonai install.

6. MCP host — serve agents to Claude Desktop / Cursor

   pip install "praisonai-mcp[all]"
   

   The heavy MCP host — full capability registry, recipes, and HTTP-stream auth — usable on its own or inside the full praisonai install.

7. Everything — the full wrapper

   pip install praisonai
   

   Transitively pulls praisonai-code, praisonai-bot, praisonai-train, praisonai-browser, praisonai-mcp, and praisonaiagents.

---

How to Choose

Start at the top: install praisonai unless you know you want a lighter footprint. Each lower tier trims dependencies at the cost of features.

---

How Tiers Interact at Runtime

The bot tier reaches wrapper-only features through a lazy bridge, so it stays installable on its own.

The bot tier never imports the wrapper directly — the bridge resolves config the same way whether you run bot-tier standalone or the full praisonai install.

---

Ownership Table

Each tier owns a clear slice and must not depend upward.

Tier	Package	Owns	Must not depend on
1	praisonaiagents	Agent, tools, memory, hooks, protocols	any wrapper
2a	praisonai-code	run/chat/code, Typer CLI, runtime, LLM	praisonai PyPI (lazy bridge only)
2b	praisonai-bot	Bots, gateway, channel CLI, OS daemon, gateway scheduler tick	praisonai PyPI (lazy bridge for jobs/UI)
2c	praisonai-train	LLM fine-tuning (Unsloth), iterative agent training, train CLI group	praisonai PyPI
2d	praisonai-browser	Extension bridge, CDP agent, Playwright engine, browser CLI	praisonai PyPI
2e	praisonai-mcp	Heavy MCP host, adapters, recipe bridge, HTTP-stream auth	praisonai PyPI (lazy _wrapper_bridge)
3	praisonai	Framework adapters (CrewAI/AutoGen), serve, dashboard, async jobs API	—

Publish order: praisonaiagents → praisonai-code + praisonai-bot + praisonai-train + praisonai-browser + praisonai-mcp → praisonai

---

Config Kernel

Shared configuration lives in praisonai_code/cli/configuration/. The praisonai-bot tier reaches it through a lazy internal bridge, so the bot package stays installable on its own without importing the wrapper. You never call this bridge directly — it resolves config the same way whether you run bot-tier standalone or the full wrapper.

---

Common Patterns

Embed agents in my app

Install the SDK and call Agent directly.

from praisonaiagents import Agent

agent = Agent(name="assistant", instructions="You are a helpful assistant.")
agent.start("Summarise the latest release notes.")

Deploy a bot with scheduled updates, no CrewAI

Install the bot tier and run a scheduled agent tick — no wrapper required.

import asyncio
from praisonaiagents import Agent
from praisonaiagents.scheduler import ScheduleRunner, FileScheduleStore
from praisonai_bot.scheduler import ScheduledAgentExecutor

agent = Agent(name="watcher", instructions="Summarise new alerts.")
runner = ScheduleRunner(FileScheduleStore("./schedules"))
executor = ScheduledAgentExecutor(runner=runner, agent_resolver=lambda _id: agent)

asyncio.run(executor.run_loop(interval=15.0))

---

Best Practices

Start with praisonai unless you know you want a lighter install

The full wrapper pulls every tier and is the recommended default. Drop to praisonai-code, praisonai-bot, praisonai-train, praisonai-browser, praisonai-mcp, or praisonaiagents only when you have a specific reason to trim dependencies.

Shims keep old imports working — you don't need to migrate immediately

After the C9 module moves, the old praisonai.scheduler.executor and praisonai.scheduler.condition_gate paths remain importable as backward-compatibility shims when the wrapper is installed. New code should import from praisonai_bot.scheduler.*, but existing code keeps running.

The scheduler tick works standalone from bot tier; RunPolicy needs the wrapper

ScheduledAgentExecutor ships in praisonai-bot, so scheduled ticks run without the wrapper. The RunPolicy safety gate stays in praisonai — install the wrapper if you need it for unattended runs.

---

Related

Installation

Six-package install comparison

praisonai-browser Package

Browser-tier package guide

praisonai-mcp Package

MCP-host package guide

praisonai-bot SDK

Bot-tier package reference

praisonai-train Package

Training-tier package guide

Scheduler Pre-Run Gate

Cost-efficiency gate for scheduled ticks

Scheduled Run Policy

Safety gate for unattended runs