Introduction

What is Amplifier?

A modular AI agent framework.
Open. Flexible. Built on kernel principles.

✓ Active

February 2026

The Problem

Most AI tools are black boxes

Vendor Lock-In

Locked to one AI provider. Can't switch. Can't compare. No leverage.

No Customization

The tool does what it does. No way to add capabilities. No extensibility.

Zero Visibility

What's happening? Why did it fail? You can't see. You can't debug.

The Solution

Amplifier

A modular AI agent framework.
Provider-agnostic. Fully extensible. Completely observable.

Design Philosophy

Built on Linux kernel principles

✓ Mechanism, not policy — Kernel provides building blocks, not dictates
✓ Tiny kernel — Core coordination layer is 1,344 lines of Python
✓ Powerful modules — 6 protocol types; all features live at the edges
✓ Event-driven — Everything emits events for full observability

Architecture

Layered composability

Applications

↓

Bundles (composition)

↓

Providers

Tools

Hooks

↓

Kernel (1,344 lines)

What Makes It Different

Key capabilities

Provider-Agnostic

Switch between Anthropic, OpenAI, Azure, Ollama, and more. Not locked to one vendor.

6 Module Types

Provider, Tool, Orchestrator, Context, Hook, and Approval — each a clean protocol interface.

Event Observability

Every action emits events. Full visibility into what's happening. Debug with confidence.

Recipe Workflows

Declarative multi-step workflows. Approval gates. Resumability. Repeatable execution.

Multi-Agent

Session forking for parallel work. Sub-agents with specialized contexts. Coordinate complex tasks.

Composable Bundles

Bundles compose capabilities. Mix and match. Build custom configurations without code.

The Ecosystem

Growing fast, staying modular

35

Repositories

451

Commits (core + foundation)

48

Bundles in ecosystem

25

Agent definitions

6

Module protocol types

5

Contributors

Use Cases

Who should use Amplifier?

→ Developers who want an AI coding assistant they can customize
→ Teams who need multi-provider flexibility
→ Power users frustrated with black-box AI tools
→ Organizations building AI-powered applications
→ Anyone curious about how modern AI agents work

Sources

Research Methodology

Data as of: February 20, 2026

Feature status: Active

Research performed:

Core commits: cd amplifier-core && git log --oneline | wc -l → 134
Foundation commits: cd amplifier-foundation && git log --oneline | wc -l → 317
Kernel LOC: wc -l session.py coordinator.py interfaces.py → 1,344 lines
Full core package: find amplifier_core -name "*.py" | xargs wc -l → 8,134 total (6,677 excl. tests)
Protocol types: grep "class.*Protocol" interfaces.py → 6 protocols
Ecosystem repos: ls ~/dev/ANext/ | grep -ci amplifier → 35
Bundle cache: find ~/.amplifier/cache -maxdepth 1 -type d | wc -l → 48 bundles
Agents: find . -name "*.md" -path "*/agents/*" | wc -l → 25 (foundation)
Contributors: git log --format="%an" | sort -u → 5 unique

Primary contributor: Brian Krabach (~97% of core commits, 130 of 134)

First core commit: October 8, 2025

Gaps: Bundle count from local cache; may differ from published registry. Commit totals are core + foundation only; other repos not counted.

Ready to start?

Complete tutorial available online

Tutorial: ramparte.github.io/amplifier-tutorial
GitHub: github.com/microsoft/amplifier

Install in 90 seconds. Start building immediately.