Buddy AI

Build AI agents that work for you — not just chat with you.

Buddy AI is a batteries-included Python framework for building, deploying, and managing intelligent agents. One unified API gives you 30+ LLM providers, 200+ tools, persistent memory, RAG knowledge, multi-agent teams — and an autonomous virtual employee that ships real work while you sleep.

pip install buddy-ai[all]

Get started Read the docs

⭐ Featured in v2.2.0

New in v2.2.0 — Built-in Prompt Caching

Save up to 90 % of repeated prompt-token costs with a single flag. Anthropic gets explicit cache_control breakpoints on system prompt, tools, and conversation history. OpenAI gets automatic server-side caching with cache-hit metrics surfaced automatically.

agent = Agent(model=Claude(id="claude-opus-4-5"), cache_prompt=True)

Prompt Caching guide

New in v2.2.0 — Competency Engine

Teaching AI Agents to Know What They Don't Know. Read the full story — plain English, runnable code, honest framing.

Read the release article

🤔 Why Buddy AI?

One API, every model

Swap between OpenAI, Anthropic, Google, AWS Bedrock, Azure, Cohere and 30+ providers without rewriting your agent. Mix and match, fail over, and integrate custom models.
Tools out of the box

Ship faster with 200+ built-in tools, and add your own with a simple decorator. Automatic schema generation means function calling just works.
Memory that persists

Conversation summarization, long-term storage, and user-specific memories that carry across sessions — so your agents actually remember.
Production from day one

FastAPI REST apps, interactive UIs, Docker & Kubernetes support, and enterprise security features built into the framework.

🚀 Key Features

Advanced Agent System
- Intelligent agents with persistent memory and personality
- Multi-agent teams with sophisticated orchestration
- Agent evolution through automated improvement
- Personality engine for behavioral modeling
Agent framework
Multi-Model Support
- 30+ LLM providers under one unified interface
- Model switching and failover
- Custom model integration
Model providers
Extensible Tool System
- 200+ built-in tools for common tasks
- Custom tool creation with simple decorators
- Function calling, composition, and chaining
Tools & functions
Knowledge & RAG
- Multi-format document processing (PDF, DOCX, MD…)
- Vector database integration (ChromaDB, Pinecone, Weaviate)
- Semantic search with intelligent ranking
Knowledge management
Teams & Workflows
- Multi-agent collaboration and communication
- Template-based workflow automation
- A robust execution engine
Team collaboration
Deploy Anywhere
- FastAPI integration for REST APIs
- Interactive apps and playgrounds
- Docker & Kubernetes ready
Deployment options

🚦 Quick Start

Spin up your first agent in a few lines:

from buddy import Agent
from buddy.models.openai import OpenAIChat

agent = Agent(
    name="Assistant",
    model=OpenAIChat(),
    instructions="You are a helpful assistant.",
)

response = agent.run("Hello, what can you do?")
print(response.content)

Add memory and a real web-search tool to build a research assistant:

from buddy import Agent
from buddy.models.openai import OpenAIChat
from buddy.tools.tavily import TavilyTools
from buddy.memory.agent import AgentMemory

agent = Agent(
    name="ResearchBot",
    model=OpenAIChat(),
    memory=AgentMemory(),
    tools=[TavilyTools()],
    instructions="You are a research assistant that can search the web.",
)

response = agent.run("What are the latest developments in AI?")
print(response.content)

Score and route your team's competency

Want to know where your agents are strong — and send each task to the right one?

from buddy.eval.competency import CompetencyEval

ev = CompetencyEval(
    domains={"reasoning": 7, "memory": 6, "deployment": 4},
    base=9,
    dependency={("reasoning", "deployment"): 0.4},
)
result = ev.run()
print(f"Competency index: {result.index:.2%}")
print(f"Weakest domain:   {result.weakest_domain()}")

Dive into the Competency Engine →

🏗️ Architecture Overview

graph TD
    A[User Input] --> B[Agent]
    B --> C[Model Provider]
    B --> D[Memory System]
    B --> E[Tool System]
    B --> F[Knowledge Base]

    C --> G[OpenAI/Anthropic/etc.]
    D --> H[Vector Storage]
    E --> I[Built-in Tools]
    E --> J[Custom Tools]
    F --> K[Documents/URLs]

    B --> L[Response Generation]
    L --> M[User Output]

🎯 Core Modules

Module	Description	Highlights
Agent	Core agent implementation	Memory, tools, personality, evolution
PULSE	Autonomous virtual employee	Real file output, persistent memory, meetings, standups, web UI
Competency Engine	Scoring & orchestration	Domain scoring, gap prioritization, runtime routing
Models	LLM provider integrations	30+ providers, unified interface
Tools	Function calling system	200+ tools, custom creation
Memory	Memory management	Conversation, long-term, user memories
Knowledge	RAG and document processing	Multi-format, vector search
Team	Multi-agent collaboration	Orchestration, communication
Workflows	Process automation	Template-based, execution engine
Token Efficiency	6-layer cost reduction	Prompt cache, proxy cache, token budget, tool trimming, history compression
Integrations	LangChain & LangGraph interop	Buddy models/agents as LangChain & LangGraph components, competency routing edges
Training	Model fine-tuning	Data prep, training, evaluation
CLI	Command-line interface	Workspace management, operations
API	REST API framework	FastAPI integration, endpoints

📚 Documentation Structure

This documentation is organized into the following sections:

Getting Started — Installation, setup, and first steps
Core Concepts — Understanding the fundamental components
Agent Framework — Deep dive into agent capabilities
Model Providers — Working with different LLM providers
Tools & Functions — Building and using tools
Memory System — Managing agent memory and state
Knowledge Management — RAG and document processing
Advanced Features — Multi-modal, reasoning, planning, competency, prompt caching
Integrations — Use Buddy with LangChain and LangGraph
Examples & Tutorials — Practical examples and use cases

🤝 Community & Support

GitHub: esasrir91/buddy-ai
Issues: Bug Reports & Feature Requests
Discussions: Community Forum

📄 License

This project is licensed under the MIT License — see the LICENSE file for details.

Ready to build intelligent AI agents? Get started now →