PraisonAI
Build and run orchestrated AI-agent workflows through Python, JavaScript, CLI, or YAML.
- Source repo
- MervinPraison/PraisonAI
- Stars
- ★ 9.3k
- Last updated
- today
- License
- MIT
- Primary language
- Python
- FA score
- 42/100 · Major gaps
At a glance
- How it runs
- Works with
- Universal · cross-platformCodex · Claude Code · OpenAI API · Claude API
- Cost
- Free software; you pay for model usage
- Setup effort
- Low · running in minutes
- You'll need
- Typical use
- A Python team building a research-to-writing pipeline can place specialized roles in
Agentsand start the team sequentially. - Not a fit if
- Users wanting zero-config use without supplying model keys and channel tokens
- Ops teams needing an official production topology or compatibility matrix
- Source review
- 42/100 · Major gaps 2 safety controls not found
What does this agent do, and when should you use it?
PraisonAI is a collection of SDK, CLI, and visual components for building AI-agent applications. Its Python core package, praisonaiagents, exposes Agent, Agents, and MCP for single-agent and multi-agent execution. The repository also documents the praisonai CLI, YAML-defined teams, a Claw Dashboard, a Langflow Flow Builder, and a lightweight chat UI. Documented capabilities include planning, workflow routing and parallelism, memory, RAG, guardrails, session persistence, and MCP-based tool connections. Tasks can be started from Python, JavaScript, the command line, or YAML; the optional Dashboard provides local management pages at port 8082.
A developer instantiates Agent(instructions=...) and calls agent.start(...) with a task, or groups agents with Agents(agents=[...]) and calls agents.start(). Agents can use Python functions decorated with @tool and connect to MCP servers; the README shows stdio, HTTP, and WebSocket MCP configurations. Workflows document route(), parallel(), loop(), and repeat() patterns alongside memory, knowledge/RAG, sessions, and database persistence. The CLI can execute YAML-defined teams and manage memory, knowledge, sessions, and MCP settings, while also exposing research, workflow, and scheduling commands. Optional commands launch a local Claw Dashboard (praisonai claw) or a Langflow-based visual flow interface (praisonai flow).
- A Python team building a research-to-writing pipeline can place specialized roles in
Agentsand start the team sequentially. - An engineering team connecting local or remote MCP services can configure tools through
MCP("npx ..."), HTTP, or WebSocket endpoints. - An operations user who wants versioned, no-code agent definitions can maintain
agents.yamland runpraisonai agents.yaml. - An application that needs persisted messages, runs, and traces can configure an
Agentwithdb(database_url=...)and asession_id. - A support team managing agents across Telegram, Discord, Slack, or WhatsApp can install the Claw extra and use its local Dashboard.
- A workflow designer who prefers a visual canvas can use Langflow’s Agent and Agent Team components for sequential or parallel flows.
How do you install or deploy this agent?
You need a Python environment capable of running pip and network access to the selected model service. Install the core SDK and set the OpenAI credential:
pip install praisonaiagents
export OPENAI_API_KEY="your-api-key"Then create a Python file that imports Agent from praisonaiagents. For the full CLI, run pip install praisonai. The documented JavaScript SDK install command is npm install praisonai.
How do you use this agent?
Minimal Python usage:
from praisonaiagents import Agent
agent = Agent(instructions="You are a senior data analyst.")
agent.start("Analyze the top 3 tech trends of 2026 and format as a markdown table.")For YAML, put framework: praisonai, a topic, and agents in agents.yaml, then run praisonai agents.yaml. For the Claw Dashboard, install pip install "praisonai[claw]", configure OPENAI_API_KEY, and additionally set TAVILY_API_KEY for its built-in web search; run praisonai claw and open http://localhost:8082.
What are this agent's strengths and limitations?
- The core surface is compact:
Agent,Agents, andMCPcover individual agents, teams, and MCP tool connections. - It offers documented Python, JavaScript, CLI, YAML, Dashboard, and Langflow delivery paths from one repository.
- The README explicitly documents MCP over stdio, HTTP, WebSocket, and SSE, with configuration examples.
- Its workflow feature set includes routing, parallel execution, loops, repeated evaluation, conditional branching, and checkpoints.
- It documents examples for multiple model providers, including OpenAI, Anthropic, Gemini, and Ollama.
- The first Python example requires
OPENAI_API_KEY; other providers bring their own credentials and service dependencies. - Claw’s built-in web search requires an additional
TAVILY_API_KEY, and channel connections require platform-token configuration. - The local MCP example uses
npx, so that route depends on the relevant Node.js/package execution environment. - The README lists many features, but the supplied material does not provide a compatibility matrix, resource requirements, or a production deployment topology for the optional components.
- Custom tools handling model- or user-supplied input require the adopter to validate and sanitize inputs; the README specifically warns against
eval(),exec(), andsubprocess.
How does this agent compare with similar options?
PraisonAI documents Claude Code, Gemini CLI, and Codex as external agents it can orchestrate. Its described role is to coordinate those tools through its own Agent, workflow, and tool interfaces, rather than replace their command-line products.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| PraisonAI This agent | 42 · Major gaps | Library / SDKFree + model costs | ★ 9.3k | today | Python | Codex · Claude Code · OpenAI API · Claude API |
| Youtu-Agent | 27 · Major gaps | CLIFree + model costs | ★ 4.6k | 7mo ago | Python | OpenAI API |
| MCP Memory Service | 68 · Some gaps | MCP serverFree | ★ 2.1k | 1d ago | Python | ChatGPT · Codex · Claude Code · Claude.ai |
| Microsoft Agent Framework Hands-On Samples | 50 · Major gaps | Library / SDKFree + model costs | ★ 390 | 5mo ago | Jupyter Notebook | — |
How does FollowAgents rate this agent?
Why each dimension lost points
Evidence shows: README emphasizes security notes (avoid eval/exec), SECURITY.md provides vulnerability reporting process, but no least privilege or user confirmation mechanisms. Data flow transparency insufficient, sensitive data handling only mentions environment variables. Dependency security has security policy but no specific audit. External effects include install script and MCP tools, but not explicitly limited. Rollback feature mentioned via shadow git checkpoints. Source attribution has MIT license and author info. Deductions: lack of user confirmation, data flow transparency, detailed sensitive data handling.
Evidence shows: README and documentation structure consistent, provides multiple use cases and examples. Dependency availability not explicitly stated, but mentions support for many LLM providers. Failure messages not detailed. Deductions: dependency availability and failure messages lack specific evidence.
Evidence shows: README clearly defines target audience (developers) and multiple use cases (research, code generation, etc.). Capability boundaries not explicit, but feature list provided. Trigger precision not detailed. Environment fit supports multiple LLMs and deployment methods. Deductions: capability boundaries and trigger precision lack clear explanation.
Evidence shows: README structure clear, provides installation instructions (pip/npm/curl), naming stable (praisonaiagents etc.), extensive examples and documentation links. Known limitations not explicitly listed. License is MIT. Version and changelog not provided. Maintenance responsibility has SECURITY.md and contribution guidelines. Deductions: known limitations, version and changelog missing.
Evidence shows: Output usability demonstrated through examples and docs, marginal value high (multi-agent, MCP, etc.), cost-benefit not explicitly stated. Deductions: cost-benefit lacks specific data.
Evidence shows: Claims in README supported by documentation links, but no independent verification. Cross-source corroboration limited. Facts and inferences not clearly separated. Deductions: lack of independent verification and fact-inference separation.
- Not found in source: confirmation before actingTurn on (or add) a confirmation step before it acts, and try it in a sandbox or test environment before real data.
- Not found in source: data-flow disclosureWatch which external services it contacts (proxy or firewall logs) and keep sensitive data out until you know where it goes.
- Install script (curl | bash) poses supply chain risk; review script content.
- MCP tools and external agents may execute arbitrary code; configure permissions carefully.
- Environment variables (e.g., API keys) handling must ensure secure storage to avoid leakage.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
FAQ
What credentials are needed for a first run?
OPENAI_API_KEY. Claw’s built-in web search also requires TAVILY_API_KEY.Can I use it without writing Python?
praisonai CLI, and YAML definitions executable with praisonai agents.yaml.Can it connect to existing tool servers?
MCP. The README shows local npx/stdio, HTTP, WebSocket, and environment-variable-based configurations.Can it retain conversations and memory?
memory=True enables file-based memory, while db(database_url=...) with session_id persists messages, runs, and traces.