0.1.0b1

---

Overview

AgentFabric is an open-source runtime layer for AI agents. It provides the foundational infrastructure — memory, tools, permissions, observability, scheduling, pipelines, and a desktop studio — that every agent needs to be production-ready.

AgentFabric is not another agent framework. It does not replace LangGraph, CrewAI, AutoGen, or the OpenAI SDK. It works alongside them, providing the infrastructure layer underneath.

┌──────────────────────────────────────────────────┐
│                YOUR APPLICATION                   │
└──────────────────┬───────────────────────────────┘
                   │
┌──────────────────▼───────────────────────────────┐
│           YOUR FRAMEWORK (optional)               │
│     LangGraph · CrewAI · AutoGen · OpenAI SDK     │
└──────────────────┬───────────────────────────────┘
                   │
╔══════════════════▼═══════════════════════════════╗
║                 AgentFabric Runtime                   ║
║  Memory · Tools · Permissions · Observability     ║
║  Scheduler · Pipelines · EventBus · Plugins       ║
║  Knowledge Graph · Workspaces · Agent Teams       ║
╚══════════════════════════════════════════════════╝

---

⚡ Getting Started

Installation

pip install agent-fabric

Create Your First Agent

from agent_fabric import Agent

agent = Agent("researcher")
result = agent.run("What are the latest breakthroughs in quantum computing?")
print(result.text)

No configuration files. No server to start. No database to set up. API keys are read from environment variables.

export OPENAI_API_KEY="sk-..."

---

📖 Core Concepts

Agents

An agent is an autonomous entity that can think, use tools, remember information, and collaborate with other agents.

from agent_fabric import Agent

agent = Agent(
    name="analyst",
    model="gpt-4o",
    system_prompt="You are a data analyst.",
)
result = agent.run("Analyze the trends in this dataset")

Tools

Tools give agents the ability to interact with the outside world. Create custom tools with the @tool decorator:

from agent_fabric import Agent, tool

@tool("Search the web for information")
def web_search(query: str) -> str:
    # Your search implementation
    return results

@tool("Read a webpage and extract its content")
def read_url(url: str) -> str:
    # Your implementation
    return content

agent = Agent("researcher", tools=[web_search, read_url])

Memory

Agents automatically persist memory across sessions. You can also use memory directly:

from agent_fabric import memory

memory.store("Project deadline is March 15th", tags=["projects", "deadlines"])
results = memory.search("upcoming deadlines")

Agent Teams

Multiple agents can collaborate on complex tasks:

from agent_fabric import Agent, Team

researcher = Agent("researcher", role="Research topics thoroughly")
writer = Agent("writer", role="Write clear, engaging content")
reviewer = Agent("reviewer", role="Review and improve quality")

team = Team(
    agents=[researcher, writer, reviewer],
    strategy="sequential",
)
result = team.run("Create a comprehensive report on AI infrastructure")

Pipelines

Define multi-step workflows as directed acyclic graphs:

# pipeline.yaml
name: morning-briefing
nodes:
  - id: scan_news
    type: skill
    skill: research
    inputs: { topic: "Top AI news today" }

  - id: scan_emails
    type: tool
    tool: gmail_read
    inputs: { filter: "is:unread" }

  - id: compile
    type: agent
    agent: writer
    depends_on: [scan_news, scan_emails]

from agent_fabric import Pipeline, Schedule

pipeline = Pipeline.from_yaml("pipeline.yaml")

# Run once
pipeline.run()

# Or schedule it
Schedule(pipeline, cron="0 8 * * *")  # Every day at 8 AM

Bring Your Own Agent

Already have an existing agent? Add AgentFabric infrastructure without rewriting it:

from agent_fabric import enhance

# Your existing agent — any Python class
class MyExistingBot:
    def run(self, task):
        return openai.chat.completions.create(...)

bot = enhance(MyExistingBot(), memory=True, observe=True)
bot.run("Analyze this data")
# Now it has persistent memory, logging, and metrics

Framework Adapters

Use AgentFabric with your existing framework:

from agent_fabric.adapters import LangGraphAdapter

agent = LangGraphAdapter(my_langgraph_app)
agent.run("Do research")
# Your LangGraph agent now has AgentFabric memory and observability

Adapters are available for OpenAI Agents SDK, LangGraph, and CrewAI.

---

🏗️ Architecture

AgentFabric is built as a modular runtime with the following core systems:

Design Principles

• Embedded by default — The SDK runs in-process. No server, no database setup, no infrastructure to manage.
• Zero configuration — Sensible defaults for everything. Configuration files are optional and only needed for customization.
• Progressive complexity — Simple use cases require simple code. Advanced capabilities are available when needed.
• Framework agnostic — Works with any agent framework or no framework at all.
• À la carte — Use the full runtime or individual components independently.
• Local first — All data stays on your machine by default. No cloud dependency.

---

🖥️ AgentFabric Studio

AgentFabric Studio is a desktop application for managing your AI infrastructure visually.

Built with Tauri and SvelteKit for a lightweight, fast, cross-platform experience.

agentfabric studio

---

⚡ Why AgentFabric? Comparison vs Competitors

---

🔌 Ecosystem

LLM Providers

AgentFabric is model-agnostic. Configure your preferred provider:

Plugins

Extend AgentFabric with integrations:

MCP Compatibility

AgentFabric supports the Model Context Protocol:

• Expose AgentFabric tools as MCP servers for use in Claude Desktop, Cursor, and other MCP clients
• Consume external MCP servers as AgentFabric tools

---

🧑‍💻 SDK Reference

Quick Reference

from agent_fabric import Agent, Team, Pipeline, Schedule
from agent_fabric import tool, memory, enhance
from agent_fabric import Runtime

Runtime Configuration

For advanced use cases, configure the runtime explicitly:

from agent_fabric import Runtime, Agent

runtime = Runtime(
    workspace="my-project",
    provider="anthropic",
    model="claude-sonnet-4",
    memory_backend="qdrant",
)

agent = Agent("analyst", runtime=runtime)

CLI

agentfabric run "Research AI trends"       # One-shot agent
agentfabric agent start <name>             # Start a persistent agent
agentfabric agent list                     # List running agents
agentfabric memory search "query"          # Search memory
agentfabric pipeline run pipeline.yaml     # Run a pipeline
agentfabric schedule create --cron "..."   # Create a schedule
agentfabric studio                         # Open the Studio
agentfabric plugin list                    # List installed plugins
agentfabric workspace list                 # List workspaces

---

📖 Documentation

---

🤝 Contributing

We welcome contributions of all kinds — bug reports, feature requests, documentation improvements, and code contributions.

Please read our Contributing Guide for development setup, coding standards, and the contribution process.

---

📄 License

AgentFabric is licensed under the Apache License 2.0.

---