What Amplifier Is

Amplifier is a modular platform for AI-assisted development. The CLI you run is a reference application built on top of a small kernel (amplifier-core) plus swappable runtime modules (providers, tools, hooks, and orchestrators).

The mental picture

flowchart TB app["Your app (CLI, web app, IDE plugin, service)"] --> cfg["Load config + select bundle"] cfg --> mount["Mount plan: orchestrator + provider + tools + hooks"] mount --> loop["Orchestrator loop runs turns"] core["amplifier-core (kernel contracts)"] --> mount foundation["amplifier-foundation (bundle system)"] --> cfg modules["Runtime modules (plugins)"] --> mount subgraph Modules providers["Providers"] tools["Tools"] hooks["Hooks"] orchestrators["Orchestrators (loops)"] contexts["Context managers"] end modules --> providers modules --> tools modules --> hooks modules --> orchestrators modules --> contexts

Why it feels like "Claude Code"

The CLI can feel similar because it includes the same genre of parts: a streaming terminal UI, a tool-calling loop, local filesystem and shell tools, session persistence, etc. The key difference is that Amplifier is designed so those behaviors are assembled from modules and bundles rather than being one fixed monolith.

Useful rule of thumb

If something "mysteriously appears" (an agent, a tool, a different streaming behavior), it's almost always because your active bundle mounted a different set of modules.

No "One True Amplifier" - It's Configured Per Session

Critical Architectural Principle

There is no single "Amplifier instance" with all features. Each session is configured with a specific bundle, which determines what that session can do.

Two sessions can have completely different capabilities:

Session A (recipes bundle): Can execute multi-step workflows, has recipe-author agent
Session B (lsp-python bundle): Can navigate code with LSP, has python-code-intel agent

Same kernel, same app, different configurations = different behaviors.

Configuration: Bundles Determine Capabilities

Amplifier uses bundles to configure what each session can do. A bundle is a markdown file with YAML frontmatter that specifies:

Which modules to load (providers, tools, hooks, orchestrators)
System prompts and agent definitions
Context files and resources
Skills directories (reusable knowledge packages)
Other bundles to include (composition)

The bundle you choose determines what that session can do. Change bundles, change capabilities. See the Bundles page for details.

Entry point vs. ecosystem

The amplifier GitHub repo is the entry point and documentation hub; it depends on amplifier-app-cli and exposes the amplifier command.
The "real" platform includes many other repos:
- amplifier-core: Kernel contracts
- amplifier-foundation: Bundle system and foundational library
- Runtime modules: Providers, tools, hooks, orchestrators, context managers
- Bundles: Pre-configured capability packages

Why that matters

If you want a different UX (web UI, background service, multi-task runner), you build a different application that hosts the same core contracts and mounts the same kinds of modules.