Installation

Set up PraisonAI in your development environment

Install PraisonAI, set one provider key, and run an Agent in your own app.

Fastest start — one line, no Python knowledge needed:

curl -fsSL https://raw.githubusercontent.com/MervinPraison/PraisonAI/main/install.sh | sh

The one-liner installs a managed, isolated CLI and adds praisonai to your PATH. Keep it current with praisonai upgrade and remove it with praisonai uninstall — see Upgrade & Uninstall. Prefer pip? That still works exactly as before — pick a package below.

from praisonaiagents import Agent

agent = Agent(instructions="Be helpful")
agent.start("Summarise the top AI news today")

Bot (gateway + channels)

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

pip install praisonai

pip install praisonai-code

pip install "praisonai-train[llm]"

pip install "praisonai-browser[all]"

pip install praisonaiagents

npm install praisonai

curl -fsSL https://raw.githubusercontent.com/MervinPraison/PraisonAI/main/install.sh | sh

Six packages, one ecosystem. pip install praisonai pulls code, bot, train, browser, and SDK. Lighter options below.

Terminal-only (smaller)

pip install praisonai-code — agentic CLI without gateway or bots

Bots + gateway only

pip install "praisonai-bot[gateway,bot]" — channel runtime without the wrapper

Full (default)

pip install praisonai — wrapper, code, bot, and SDK together

Training only

pip install "praisonai-train[llm]" — LLM fine-tuning + agent training

Browser automation

pip install "praisonai-browser[all]" — Chrome extension + Playwright engine

pip install praisonai transitively installs praisonai-code, praisonai-bot, praisonai-train, praisonai-browser, and praisonaiagents. You do not need to install them separately.

One-line installer (recommended for CLI users)

If you want the CLI end-to-end, the one-line installer provisions an isolated managed environment (uv tool → pipx) and wires your PATH — no changes to your global Python:

curl -fsSL https://raw.githubusercontent.com/MervinPraison/PraisonAI/main/install.sh | sh

It is idempotent and safe to re-run. pip install praisonai is untouched — keep using it if you embed PraisonAI as a Python library.

Environment-variable overrides:

Variable	Purpose
PRAISONAI_VERSION=x.y.z	Pin a specific version (default: latest)
PRAISONAI_INSTALLER=uv|pipx	Force a specific tool manager
PRAISONAI_NONINTERACTIVE=1	Never prompt — suitable for CI

After installing, configure a provider key:

praisonai setup

Keep the managed install current with praisonai upgrade, or remove it cleanly with praisonai uninstall.

Installing PraisonAI

PraisonAI is published as six PyPI packages. Pick the one that matches what you want to do:

Full (praisonai)

The complete package — CLI, gateway, bots, integrations, YAML-driven multi-bot orchestration. Pulls praisonai-code and praisonaiagents automatically.

For gateway hot reload (event-driven gateway.yaml watching via watchdog), install the gateway extra: pip install "praisonai[gateway]".

[claw] extra: pip install "praisonai[claw]" installs the full wrapper plus praisonai-bot[gateway,bot] — the WebSocket gateway and Telegram/Discord/Slack channel bots. Use this when you want the full wrapper and all channel runtime features in one command. For gateway + bots without the wrapper, use pip install "praisonai-bot[gateway,bot]" directly (see Standalone Bot Gateway).

Sandbox extras: the sandbox backends live in the standalone praisonai-sandbox package. Install a cloud backend with pip install "praisonai[docker]", "praisonai[ssh]", "praisonai[modal]", "praisonai[e2b]", or pip install "praisonai-sandbox[daytona]" for the Daytona backend (needs DAYTONA_API_KEY). See Sandbox Backends for the full matrix.

1. Create Virtual Environment (Optional)

   Mac/Linux

   python -m venv praisonai-env
   source praisonai-env/bin/activate
   

   python -m venv praisonai-env
   .\praisonai-env\Scripts\activate
   
2. Install

   Terminal

   pip install praisonai
   
3. Configure Environment

   Set your API key. PraisonAI auto-detects whichever provider credential is present:

   OpenAI

   export OPENAI_API_KEY="${OPENAI_API_KEY:?Set OPENAI_API_KEY in your shell}"
   

   Anthropic

   export ANTHROPIC_API_KEY="${ANTHROPIC_API_KEY:?Set ANTHROPIC_API_KEY in your shell}"
   

   Gemini

   export GEMINI_API_KEY="${GEMINI_API_KEY:?Set GEMINI_API_KEY in your shell}"
   

   Groq

   export GROQ_API_KEY="${GROQ_API_KEY:?Set GROQ_API_KEY in your shell}"
   

   Ollama

   export OLLAMA_HOST=http://localhost:11434
   

   If you only set ANTHROPIC_API_KEY (or GEMINI_API_KEY, GOOGLE_API_KEY, GROQ_API_KEY, COHERE_API_KEY, or OLLAMA_HOST), praisonai run picks the matching provider’s default model automatically — no --model flag required. See Provider Auto-Detection for the full lookup table.

   Running Ollama on the default 127.0.0.1:11434? Skip export OLLAMA_HOST — PraisonAI auto-detects a local endpoint when no cloud key is set. See Keyless local-first fallback.

Code (praisonai-code)

The standalone agent runtime — run, chat, code, and the full CLI backend, without the gateway/bot integrations. Depends only on praisonaiagents.

1. Create Virtual Environment (Optional)

   Mac/Linux

   python -m venv praisonai-env
   source praisonai-env/bin/activate
   

   python -m venv praisonai-env
   .\praisonai-env\Scripts\activate
   
2. Install

   Terminal

   pip install praisonai-code
   
3. Configure Environment

   export OPENAI_API_KEY="${OPENAI_API_KEY:?Set OPENAI_API_KEY in your shell}"
   

   Any supported provider key works. See Provider Auto-Detection.

4. Run an agent

   from praisonaiagents import Agent
   
   agent = Agent(name="assistant", instructions="Be helpful")
   response = agent.start("Summarise the top AI news today")
   print(response)
   

Bot (praisonai-bot)

Standalone bots and gateway — no full wrapper. Depends on praisonaiagents only at PyPI level.

1. Install

   Terminal

   pip install "praisonai-bot[gateway,bot]"
   
2. Configure Environment

   export OPENAI_API_KEY="${OPENAI_API_KEY:?Set OPENAI_API_KEY in your shell}"
   export TELEGRAM_BOT_TOKEN="${TELEGRAM_BOT_TOKEN:?Set TELEGRAM_BOT_TOKEN in your shell}"
   
3. Start gateway

   Terminal

   praisonai-bot gateway start --host 127.0.0.1 --port 8765
   

See the praisonai-bot SDK page for Python API details.

Train (praisonai-train)

Standalone training tier — LLM fine-tuning via Unsloth and iterative agent training, without the full wrapper. Depends on praisonaiagents only at PyPI level.

As of praisonai-train 2026.7+, the [llm] extra requires CUDA 12.4+ and installs the modern Unsloth / TRL / torch 2.6+ stack. The install command is unchanged.

1. Create Virtual Environment (Optional)

   Mac/Linux

   python -m venv praisonai-env
   source praisonai-env/bin/activate
   

   python -m venv praisonai-env
   .\praisonai-env\Scripts\activate
   
2. Install

   Terminal

   pip install "praisonai-train[llm]"
   

   LLM fine-tuning needs Python 3.11, CUDA 12.4, and torch>=2.6 (plus unsloth>=2025.9.1, trl>=0.18.2, transformers>=4.51.3). Agent training (pip install praisonai-train) has no such requirement.

3. Configure Environment

   export OPENAI_API_KEY="${OPENAI_API_KEY:?Set OPENAI_API_KEY in your shell}"
   

   Any supported provider key works. See Provider Auto-Detection.

4. Run the trainer

   Terminal

   praisonai-train --help
   

See the praisonai-train Package page for fine-tuning and agent-training details.

Browser (praisonai-browser)

Standalone browser tier — the bridge server, CDP agent, and Playwright engine for Chrome automation, without the full wrapper. Depends on praisonaiagents only at PyPI level.

1. Create Virtual Environment (Optional)

   Mac/Linux

   python -m venv praisonai-env
   source praisonai-env/bin/activate
   

   python -m venv praisonai-env
   .\praisonai-env\Scripts\activate
   
2. Install

   Terminal

   pip install "praisonai-browser[all]"
   
3. Configure Environment

   export OPENAI_API_KEY="${OPENAI_API_KEY:?Set OPENAI_API_KEY in your shell}"
   

   Any supported provider key works. See Provider Auto-Detection.

4. Run a goal

   Terminal

   praisonai-browser run "Go to google and search praisonai"
   

See the praisonai-browser Package page and the Browser CLI for the full command set.

Agents (SDK only)

The pure Python SDK — no CLI, no gateway, no heavy dependencies. Import directly in your own application.

1. Create Virtual Environment (Optional)

   Mac/Linux

   python -m venv praisonai-env
   source praisonai-env/bin/activate
   

   python -m venv praisonai-env
   .\praisonai-env\Scripts\activate
   
2. Install

   Terminal

   pip install praisonaiagents
   
3. Configure Environment

   export OPENAI_API_KEY="${OPENAI_API_KEY:?Set OPENAI_API_KEY in your shell}"
   
4. Use in your app

   from praisonaiagents import Agent
   
   agent = Agent(name="assistant", instructions="Be helpful")
   response = agent.start("Hello!")
   print(response)
   

No Code

1. Create Virtual Environment (Optional)

   Mac/Linux

   python -m venv praisonai-env
   source praisonai-env/bin/activate
   

   python -m venv praisonai-env
   .\praisonai-env\Scripts\activate
   
2. Install PraisonAI

   pip install praisonai
   
3. Set API Key

   Set your API key for the provider you want to use. PraisonAI auto-detects whichever credential is present:

   export OPENAI_API_KEY="${OPENAI_API_KEY:?Set OPENAI_API_KEY in your shell}"
   

   Use Anthropic, Gemini, Groq, or Ollama instead? Just set that provider’s key and praisonai run picks the right model automatically. See Provider Auto-Detection.

JavaScript

1. Install PraisonAI

   npm install praisonai
   
2. Set API Key

   export OPENAI_API_KEY="${OPENAI_API_KEY:?Set OPENAI_API_KEY in your shell}"
   

TypeScript

1. Install PraisonAI

   npm install praisonai
   
2. Set API Key

   export OPENAI_API_KEY="${OPENAI_API_KEY:?Set OPENAI_API_KEY in your shell}"
   

All existing praisonai CLI verbs (run, chat, code, gateway, bot, …) continue to work unchanged — the wrapper routes them through the praisonai-code runtime. You do not need to change any scripts.

Package dependency chain

Each layer depends on the one to its left. Installing praisonai pulls all three; installing praisonai-code pulls only praisonaiagents; installing praisonaiagents has no PraisonAI dependencies.

Generate your OpenAI API key from OpenAI You can also use other LLM providers like Anthropic, Google, etc. Please refer to the Models for more information.

Which Install Do You Need?

Terminal-only (smaller)

pip install praisonai-code — agentic CLI: run, chat, code, runtime. No bots or gateway.

Full (default)

pip install praisonai — everything: bots, gateway, kanban, dashboard, and the agentic CLI.

pip install praisonai transitively installs praisonai-code and praisonaiagents. You do not need to install them separately.

How It Works

Pip installs the package, you export one provider key, and the CLI or SDK resolves the matching model at run time.

Best Practices

Install only what you need

Use praisonaiagents for a pure Python SDK, praisonai-code for the terminal runtime, and praisonai for the full wrapper with bots and gateway.

Use a virtual environment

Create a venv before installing to keep PraisonAI isolated from your system Python.

Export keys, don't hard-code them

Set OPENAI_API_KEY (or your provider’s key) in your shell or .env. Never inline the raw value in code.

Skip --model on first runs

PraisonAI auto-detects the provider from whichever key is present and picks a sensible default model.

Related

Quick Start

Build your first AI agent.

Models

Choose a provider and model.

---

Quick Install

The one-liner installer uses an isolated tool manager (uv tool → pipx → bootstrap uv) and exposes praisonai via a ~/.local/bin/praisonai shim — your global Python environment is untouched. See Isolation Backends for details.

macOS / Linux / WSL

curl -fsSL https://raw.githubusercontent.com/MervinPraison/PraisonAI/main/install.sh | sh

Windows

iwr -useb https://praison.ai/install.ps1 | iex

Installs praisonai[all] into %USERPROFILE%\.praisonai\venv, wires PATH, and runs praisonai setup on first install. See Installer Internals for parameters and env-var overrides.

To install with specific extras only:

& ([scriptblock]::Create((iwr -useb https://praison.ai/install.ps1))) -Extras "ui,chat"

Alternative (no installer): pipx install praisonai or pip install "praisonai[all]".

Windows terminals: PraisonAI automatically detects legacy Windows code pages (CP1252, CP850, etc.) and falls back to ASCII-safe output. For full emoji and box-drawing rendering, switch your terminal to UTF-8:

PowerShell

$env:PYTHONIOENCODING='utf-8'
chcp 65001

set PYTHONIOENCODING=utf-8
chcp 65001

The installer detects your platform, picks a tool manager (uv tool → pipx → bootstrap uv), installs praisonai into an isolated environment, and drops a ~/.local/bin/praisonai shim on your PATH.

Requirements

• Python 3.10 or higher
• macOS, Linux, or WSL (the one-liner is POSIX; on Windows use pipx/pip)

Manage Your Install

Upgrade

praisonai upgrade — update the CLI in place, or --check for a newer version

Uninstall

praisonai uninstall — remove the managed install cleanly

---

Package Structure

PraisonAI ships as three separate PyPI packages. Understanding which to install avoids surprises at runtime.

Install command	What you get	Agentic CLI (run, chat, code, …)	Bot / gateway / pairing
pip install praisonai	Wrapper + code + agents	✅ Full	✅ Yes
pip install praisonai-code	Terminal-native agent CLI + LLM runtime	✅ Default run "…", `run —output plain	verbose	silent	actions	json	stream	stream-json, daemon, doctor, config, version—chatandcode` need the wrapper	❌ No — needs pip install praisonai
pip install praisonaiagents	Core SDK — Python API only	❌ No CLI	N/A

Standalone pip install praisonai-code supports the actions/daemon path as of PraisonAI 4.6.123+ (C7 + C9 four-tier). Use praisonai-code run --output actions|json|stream|stream-json "your prompt" for the standalone in-process Agent path, and praisonai-code daemon start --background (spawns python -m praisonai_code.runtime) for the warm runtime. run --help, config, doctor, and version also work standalone.

chat and code require the wrapper (pip install praisonai) and print a clear install hint if invoked standalone (for example, Error: chat requires the praisonai wrapper. Install the full wrapper: pip install praisonai). Default run "prompt" and all --output modes now route through the in-process Agent on a standalone install (PR #2853). Wrapper-only features (bot, claw, dashboard, gateway, identity, kanban, onboard, pairing, framework adapters, YAML --framework crewai/autogen) still require pip install praisonai — they degrade gracefully rather than raising raw import errors. See Standalone LLM Modules and the praisonai-code CLI.

The stub command groups agents, workflow, registry, memory, skills, hooks, rules, eval, package, templates, todo, research, commit, and call show a single-line install hint (<group> requires the full wrapper. Install the full wrapper: pip install praisonai) and exit 1 on standalone (PR #2854). Their native standalone-safe subcommands (memory learn, skills bundle/check/eligible, eval list-judges, eval list) are unaffected. See the praisonai-code CLI reference.

PyPI publish order

Packages are published in dependency order:

1. praisonaiagents  →  2. praisonai-code + praisonai-bot + praisonai-train + praisonai-browser  →  3. praisonai

If you pin versions, ensure all six packages resolve to the same release cycle.

Since 4.6.104, praisonai pins praisonai-code>=0.0.2, so pip install praisonai always pulls a compatible runtime automatically. See praisonaiagents SDK page for the release-cycle notes.