Amplifier Ecosystem¶
Amplifier's modular architecture allows you to mix and match capabilities. This page catalogs all available components — core infrastructure, applications, libraries, bundles, and runtime modules.
Core Infrastructure¶
The foundational kernel that everything builds on.
| Component | Description | Repository |
|---|---|---|
| amplifier-core | Ultra-thin kernel for modular AI agent system | amplifier-core |
Applications¶
User-facing applications that compose libraries and modules.
| Component | Description | Repository |
|---|---|---|
| amplifier | Main Amplifier project and entry point — installs amplifier-app-cli via uv tool install | amplifier |
| amplifier-app-cli | Reference CLI application implementing the Amplifier platform | amplifier-app-cli |
| amplifier-app-log-viewer | Web-based log viewer for debugging sessions with real-time updates | amplifier-app-log-viewer |
| amplifier-app-benchmarks | Benchmarking and evaluating Amplifier | amplifier-app-benchmarks |
| amplifier-app-actions | AI-powered GitHub Actions for automated issue triage and PR review | amplifier-app-actions |
| amplifier-eval-harness | Configurable test harness for evaluating Amplifier bundles across scenarios and providers in isolated Digital Twin Universe environments. Captures wall time, LLM time, per-call tokens, cache traffic, request counts, and per-provider/model breakdowns across the full session tree (parent + every sub-session) for cross-bundle and cross-provider comparison. | amplifier-eval-harness |
| amplifierd | Localhost HTTP daemon exposing amplifier-core and amplifier-foundation over REST and SSE — drive sessions from any language or framework | amplifierd |
| amplifier-chat | Chat UI plugin for amplifierd — browser-based conversational interface for creating and managing Amplifier sessions | amplifier-chat |
| amplifier-voice | Voice plugin for amplifierd — WebRTC voice interface using the OpenAI Realtime API, standalone or as a plugin | amplifier-voice |
Note: When you install amplifier, you get amplifier-app-cli as the executable application. amplifierd is a separate daemon that exposes Amplifier capabilities over HTTP, and amplifier-chat and amplifier-voice are plugins that extend it with web-based chat and voice interfaces.
Documentation & Learning¶
These are the canonical documentation and learning sites for the Amplifier ecosystem. They reflect the framework but do not run it.
| Component | Description | Repository |
|---|---|---|
| amplifier-app-learn | Interactive learning curriculum for Amplifier — six narrative levels, hands-on approach guides, validated architecture diagrams, and a chat assistant with live DOT rendering. React 19 + Vite SPA. | amplifier-app-learn |
Libraries¶
Foundational libraries used by applications (not used directly by runtime modules).
| Component | Description | Repository |
|---|---|---|
| amplifier-foundation | Foundational library for bundles, module resolution, and shared utilities | amplifier-foundation |
Architectural Boundary: Libraries are consumed by applications (like amplifier-app-cli). Runtime modules only depend on amplifier-core and never use these libraries directly.
Bundles¶
Composable configuration packages that combine providers, behaviors, agents, and context into reusable units.
| Bundle | Description | Repository |
|---|---|---|
| amplifier-tester | Validates Amplifier ecosystem changes (core, modules, bundles, foundation, app-cli) in isolated Digital Twin Universe environments — dynamically generates profiles, mirrors repos to Gitea, and runs targeted validation checks | amplifier-bundle-amplifier-tester |
| recipes | Multi-step AI agent orchestration with behavior overlays and standalone options | amplifier-bundle-recipes |
| browser-tester | Browser automation and testing with 3 specialized agents (operator, researcher, visual documenter) using agent-browser CLI | amplifier-bundle-browser-tester |
| context-managed | LLM-powered rolling context summarization with persistent transcript, budget-aware tracking, and on-demand history recovery via bundled read_transcript tool | amplifier-bundle-context-managed |
| design-intelligence | Comprehensive design intelligence with 7 specialized agents, design philosophy framework, and knowledge base | amplifier-bundle-design-intelligence |
| digital-twin-universe | On-demand isolated environments from declarative profiles — Incus containers with URL rewriting, PyPI overrides, and LLM API passthrough for evidence-based verification of deployed software | amplifier-bundle-digital-twin-universe |
| dot-graph | DOT/Graphviz infrastructure — knowledge, validation, rendering, and graph intelligence for the Amplifier ecosystem | amplifier-bundle-dot-graph |
| evaluation | One-stop-shop for evaluating AI agents, bundles, and recipes across the Amplifier ecosystem — provides an evaluation mode for designing evaluations plus a Python harness for running pre-defined tasks against agents in a Digital Twin Universe | amplifier-bundle-evaluation |
| execution-environments | Instance-based execution environments — create, target, and destroy local, Docker, and SSH environments on demand with 11 tools, composable wrappers, and NLSpec-aligned protocol | amplifier-bundle-execution-environments |
| foreman | Assistant pattern where the conversation assistant manages a fleet of other assistants, each with their own sessions, leveraging capabilities from amplifier-bundle-orchestration | amplifier-bundle-foreman |
| gitea | On-demand ephemeral Gitea instances for isolated git workflows — mirror repos from GitHub, work freely, and promote results back when ready | amplifier-bundle-gitea |
| issues | Persistent issue tracking with dependency management, priority scheduling, and session linking for autonomous work | amplifier-bundle-issues |
| llm-wiki | Karpathy LLM Wiki pattern as composable workflow modes — wiki-init, wiki-ingest, wiki-lint, wiki-publish, wiki-query; zero-cost-when-dormant via per-mode shared orientation | amplifier-bundle-llm-wiki |
| lsp | Core Language Server Protocol support for code intelligence operations | amplifier-bundle-lsp |
| lsp-python | DEPRECATED — forwarding stub, use python-dev instead | amplifier-bundle-lsp-python |
| lsp-rust | DEPRECATED — forwarding stub, use rust-dev instead | amplifier-bundle-lsp-rust |
| lsp-typescript | DEPRECATED — forwarding stub, use typescript-dev instead | amplifier-bundle-lsp-typescript |
| notify | Desktop and push notifications when assistant turns complete — works over SSH, supports ntfy.sh for mobile | amplifier-bundle-notify |
| observers | Orchestration pattern where background observer sessions are configured and run in the background, in parallel to provide the main session with actionable observations | amplifier-bundle-observers |
| orchestration | Adds event-driven orchestration primitives (bundle spawning, events, triggers) for multi-session coordination | amplifier-bundle-orchestration |
| python-dev | Comprehensive Python development tools - code quality (ruff, pyright), LSP integration, and expert agent | amplifier-bundle-python-dev |
| reality-check | Intent-driven verification of built software — derives acceptance tests from user conversations, deploys in a Digital Twin Universe environment, runs app specific validations, and produces gap analysis reports | amplifier-bundle-reality-check |
| routing-matrix | Declarative model routing with 13 semantic roles, 7 curated matrices, and CLI tooling — agents declare what they do, matrices resolve them to the right model | amplifier-bundle-routing-matrix |
| rust-dev | Comprehensive Rust development tools — code quality (cargo fmt, clippy, cargo check), LSP integration, and expert agent | amplifier-bundle-rust-dev |
| shadow | OS-level sandboxed environments for testing local Amplifier ecosystem changes safely | amplifier-bundle-shadow |
| skills | Skills tool and Microsoft-curated skills collection with two composable behaviors (full with curated skills, or tool-only) | amplifier-bundle-skills |
| stories | Autonomous storytelling engine with 11 specialist agents, 4 output formats (HTML, Excel, Word, PDF), and automated recipes for case studies, release notes, and weekly digests | amplifier-bundle-stories |
| superpowers | TDD-driven development workflows with brainstorm, plan, execute, verify, and finish modes — includes specialized agents and a full development cycle recipe | amplifier-bundle-superpowers |
| terminal-tester | Terminal application testing and inspection with 3 specialist agents (operator, visual-tester, debugger) using dual-mode capture — screen-dump for Ratatui/crossterm, PTY/pyte for any terminal app | amplifier-bundle-terminal-tester |
| ts-dev | Comprehensive TypeScript/JavaScript development tools - code quality, LSP, and expert agent | amplifier-bundle-ts-dev |
Usage: Bundles are loaded via the amplifier bundle commands:
# Add a bundle to the registry (name auto-derived from bundle metadata)
amplifier bundle add git+https://github.com/microsoft/amplifier-bundle-recipes@main
# Use a bundle by name
amplifier bundle use foundation
amplifier bundle use recipes
# Show current bundle
amplifier bundle current
# Check for bundle updates
amplifier bundle update --check
# Update bundle to latest
amplifier bundle update
Runtime Modules¶
Modules are loaded dynamically at runtime based on bundle configuration.
Orchestrators¶
Control the AI agent execution loop.
| Module | Description | Repository |
|---|---|---|
| loop-basic | Standard sequential execution — simple request/response flow | amplifier-module-loop-basic |
| loop-streaming | Real-time streaming responses with extended thinking support | amplifier-module-loop-streaming |
| loop-events | Event-driven orchestrator with hook integration | amplifier-module-loop-events |
Providers¶
Connect to AI model providers.
| Module | Description | Repository |
|---|---|---|
| provider-anthropic | Anthropic Claude integration (Sonnet 4.5, Opus, etc.) | amplifier-module-provider-anthropic |
| provider-openai | OpenAI GPT integration | amplifier-module-provider-openai |
| provider-openai-chatgpt | ChatGPT subscription backend (Plus/Pro/Team/Enterprise) via Codex CLI OAuth | amplifier-module-provider-openai-chatgpt |
| provider-azure-openai | Azure OpenAI with managed identity support | amplifier-module-provider-azure-openai |
| provider-chat-completions | OpenAI Chat Completions wire-format integration (llama.cpp, vLLM, LM Studio, LocalAI, etc.) | amplifier-module-provider-chat-completions |
| provider-gemini | Google Gemini integration with 1M context and thinking | amplifier-module-provider-gemini |
| provider-vllm | vLLM server integration for self-hosted models | amplifier-module-provider-vllm |
| provider-ollama | Local Ollama models for offline development | amplifier-module-provider-ollama |
| provider-github-copilot | GitHub Copilot models via the Copilot SDK | amplifier-module-provider-github-copilot |
| provider-mock | Mock provider for testing without API calls | amplifier-module-provider-mock |
Tools¶
Extend AI capabilities with actions.
| Module | Description | Repository |
|---|---|---|
| tool-filesystem | File operations (read, write, edit, list, glob) | amplifier-module-tool-filesystem |
| tool-bash | Shell command execution | amplifier-module-tool-bash |
| tool-web | Web search and content fetching | amplifier-module-tool-web |
| tool-search | Code search capabilities (grep/glob) | amplifier-module-tool-search |
| tool-task | Agent delegation and sub-session spawning | amplifier-module-tool-task |
| tool-todo | AI self-accountability and todo list management | amplifier-module-tool-todo |
| tool-skills | Load domain knowledge from skills following the Anthropic Skills format | amplifier-module-tool-skills |
| tool-mcp | Model Context Protocol integration for MCP servers | amplifier-module-tool-mcp |
| tool-slash-command | Extensible slash command system with custom commands defined as Markdown files | amplifier-module-tool-slash-command |
Context Managers¶
Manage conversation state and history.
| Module | Description | Repository |
|---|---|---|
| context-simple | In-memory context with automatic compaction | amplifier-module-context-simple |
| context-persistent | File-backed persistent context across sessions | amplifier-module-context-persistent |
Hooks¶
Extend lifecycle events and observability.
| Module | Description | Repository |
|---|---|---|
| hooks-logging | Unified JSONL event logging to per-session files | amplifier-module-hooks-logging |
| hooks-redaction | Privacy-preserving data redaction for secrets/PII | amplifier-module-hooks-redaction |
| hooks-approval | Interactive approval gates for sensitive operations | amplifier-module-hooks-approval |
| hooks-backup | Automatic session transcript backup | amplifier-module-hooks-backup |
| hooks-explanatory | Inject explanatory output style with educational ★ Insight blocks | amplifier-module-hooks-explanatory |
| hooks-streaming-ui | Real-time console UI for streaming responses | amplifier-module-hooks-streaming-ui |
| hooks-status-context | Inject git status and datetime into agent context | amplifier-module-hooks-status-context |
| hooks-todo-reminder | Inject todo list reminders into AI context | amplifier-module-hooks-todo-reminder |
| hooks-scheduler-cost-aware | Cost-aware model routing for event-driven orchestration | amplifier-module-hooks-scheduler-cost-aware |
| hooks-scheduler-heuristic | Heuristic-based model selection scheduler | amplifier-module-hooks-scheduler-heuristic |
| hook-shell | Shell-based hooks with Claude Code format compatibility | amplifier-module-hook-shell |
Using Modules¶
In Bundles¶
Command Line¶
amplifier module add tool-web --source git+https://github.com/microsoft/amplifier-module-tool-web@main
amplifier module list
amplifier module show tool-filesystem
Module Contracts¶
All runtime modules follow the same pattern and implement one of five contract types:
| Module Type | Contract | Purpose |
|---|---|---|
| Provider | PROVIDER_CONTRACT.md | LLM backend integration |
| Tool | TOOL_CONTRACT.md | Agent capabilities |
| Hook | HOOK_CONTRACT.md | Lifecycle observation and control |
| Orchestrator | ORCHESTRATOR_CONTRACT.md | Agent loop execution strategy |
| Context | CONTEXT_CONTRACT.md | Conversation memory management |
Quick Start Pattern¶
# 1. Implement the Protocol from interfaces.py
class MyModule:
pass # implement required methods
# 2. Provide mount() function
async def mount(coordinator, config):
instance = MyModule(config)
await coordinator.mount("category", instance, name="my-module")
return instance
# 3. Register entry point in pyproject.toml
# [project.entry-points."amplifier.modules"]
# my-module = "my_package:mount"
Community Modules¶
Providers¶
| Module | Description | Author | Repository |
|---|---|---|---|
| provider-bedrock | AWS Bedrock integration with cross-region inference support for Claude models | @brycecutt-msft | amplifier-module-provider-bedrock |
| provider-perplexity | Perplexity AI integration for chat completions with sonar models | @colombod | amplifier-module-provider-perplexity |
| provider-openai-realtime | OpenAI Realtime API for native speech-to-speech interactions | @robotdad | amplifier-module-provider-openai-realtime |
Tools¶
| Module | Description | Author | Repository |
|---|---|---|---|
| youtube | YouTube download, search, and account feed access — audio/video download, keyword and filtered search (YouTube Data API + yt-dlp fallback), watch history, subscriptions, liked videos, and more | @robotdad | amplifier-youtube |
| tool-whisper | Speech-to-text transcription using OpenAI's Whisper API | @robotdad | amplifier-module-tool-whisper |
| tool-rlm | Recursive Language Model (RLM) for processing 10M+ token contexts via sandboxed Python REPL | @michaeljabbour | amplifier-module-tool-rlm |
Hooks¶
| Module | Description | Author | Repository |
|---|---|---|---|
| hooks-concise-display | Cleaner, more condensed terminal output for Amplifier sessions | @obra | amplifier-module-hooks-concise-display |
| hooks-compact | Compress verbose bash stdout via command-aware filters (git, pytest, cargo, npm, ruff, etc.) before it enters the LLM context | @samueljklee | amplifier-module-hooks-compact |
SECURITY WARNING: Community modules execute arbitrary code in your environment with full access to your filesystem, network, and credentials. Only use modules from sources you trust. Review code before installation.