PraisonAI Documentation

PraisonAI telemetry is designed to be privacy-first and anonymous.

What is collected:

• Anonymous session IDs (no user identification)
• Event counts (agent executions, task completions)
• Success/failure rates
• Tool usage patterns

What is NOT collected:

• User content (prompts, responses, or data)
• Personal information (IPs are anonymized)
• Sensitive data (API keys, passwords)

To completely opt-out of all tracking, set the environment variable DO_NOT_TRACK=true or PRAISONAI_DISABLE_TELEMETRY=true.

The Python SDK is organized into three distinct tiers. Depending on your needs, you may interact with different packages:

• praisonaiagents (Core SDK): Contains the fundamental building blocks including Agent, tools, memory, hooks, and protocols.
• praisonai-code (Terminal CLI): Provides the command-line interface for run, chat, and code operations.
• praisonai (Wrapper): Acts as a gateway and includes multi-bot orchestration (BotOS), framework adapters (e.g., CrewAI, AutoGen), and wrapper CLI commands like bot, gateway, and pairing.

PraisonAI offers several packages depending on your development needs:

• Core SDK (praisonaiagents): For pure Python development. pip install praisonaiagents
• CLI (praisonai): For terminal-based developers. pip install praisonai
• Claw Dashboard: Connect agents to Telegram, Slack, or Discord. pip install "praisonai[claw]"
• Flow Visual Builder: Drag-and-drop workflow creation. pip install "praisonai[flow]"
• PraisonAI UI: A clean chat interface. pip install "praisonai[ui]"
• JavaScript SDK: For JS/TS environments. npm install praisonai

PraisonAI supports various agent architectures to suit different complexity levels:

• Single Agent: A standalone agent performing specific tasks.
• Multi Agents: Multiple agents collaborating to solve complex problems.
• Auto Agents: Agents that can autonomously determine their own steps.
• Self Reflection AI Agents: Agents that review and refine their own outputs.
• Reasoning AI Agents: Agents designed for deep logical extraction and reasoning.
• Multi Modal AI Agents: Agents capable of processing multiple types of input (e.g., text and images).

PraisonAI automatically tracks agent executions and task completions. The following metrics are collected and sent to PostHog upon program exit:

• Agent executions: Total number of times an agent was run.
• Task completions: Total number of tasks successfully completed within a workflow.
• Errors: Count of errors encountered during execution.
• Session ID: A unique identifier for the current session.

PraisonAI provides three distinct MCP-related layers. It is important not to conflate them:

1. Client Layer (praisonaiagents[mcp]): Used to connect agents to external MCP servers.
2. Light Server Layer (praisonai-code): Provides a basic ToolsMCPServer via the command praisonai serve mcp.
3. Heavy Host Layer (praisonai-mcp): This package. It provides a full capability/recipe MCP server.

PraisonAI agents have been optimized for faster initialization and execution, particularly for simple tasks like agent.start("..."). These optimizations are transparent to the user and maintain 100% backward compatibility with all existing public APIs.

Key performance benefits include:

• Initialization time: ~50% faster.
• First response time: ~30% faster.
• Memory usage: ~40% lower.

Optimizations include lazy loading of the rich console, system prompt caching, tool formatting caching, deferred knowledge processing, one-time logging configuration, and lazy agent ID generation.

The PraisonAI LLM class includes several internal performance optimizations designed to speed up agent initialization and tool processing. These optimizations are automatic and transparent to the user, maintaining 100% backward compatibility with existing public APIs.

Key performance features include:

• Fast Initialization: Subsequent LLM instances are ~3.4x faster due to one-time logging configuration.
• Efficient Tool Processing: Tool formatting is up to 1764x faster on cache hits using an internal _formatted_tools_cache.
• Memory Management: An internal _max_cache_size of 100 prevents unbounded memory growth during long-running sessions.
• Lazy Loading: The console object is only instantiated when accessed, saving initialization time when verbose=False.

Use the praisonai-browser CLI to start the browser service, execute automation tasks, or run diagnostic checks.

1. Start the browser service on a specific port:

praisonai-browser start --port 8765

2. Run an automation task with a natural language goal:

praisonai-browser run "Search for AI news on Google"

3. Run diagnostics to check the environment:

praisonai-browser doctor

Note: If using the full praisonai stack, use the commands praisonai browser start and praisonai browser run "..." instead.