## The Problem

If you build software with AI today, you probably use more than one tool. Claude for architecture,
Codex for implementation, Copilot for in-editor completions, Gemini for research. Each tool has its
own session history, its own context, its own API. Nothing connects.

And if you're building AI-powered software — an agent, an integration, a workflow — you're writing
the same plumbing everyone else writes: provider adapters, streaming logic, tool execution, session
management, storage. When a provider changes its API, thousands of projects fix the same breakage
independently.

Makaio exists because this work is infrastructure, and infrastructure belongs in a shared layer.

## The Makaio Ecosystem

Makaio is an ecosystem of open-source projects for AI-tool orchestration. At its core is the
**Makaio Framework** — a typed, bus-centric runtime that connects providers, tools, and sessions.
Around it sit **SDKs** that bring the bus protocol to any language, and a growing set of
community-contributed adapters and extensions.

This documentation covers the Makaio Framework — referred to simply as "Makaio" throughout these
docs.

## The Makaio Framework

Makaio is a typed, bus-centric framework for orchestrating AI agents, tools, and sessions. It is
not a provider SDK (there are [good ones](https://ai-sdk.dev/)). It is not an agent-building
framework (there are [good ones](https://langchain-ai.github.io/langgraph/) too). It sits above
both:

- **Provider SDKs** solve "call any model through one API"
- **Agent frameworks** solve "build a new AI agent system from scratch"
- **Makaio** solves "connect the AI tools you already use, share session context across them, and
  stop rebuilding the same plumbing"

These layers are complementary. A Makaio adapter can use a provider SDK internally. An agent
framework could participate in the Makaio bus as an extension. The layers compose — they don't
compete.

## Shared Infrastructure, Not Reinvented Wheels

Every project that integrates an AI provider maintains its own adapter code, its own streaming
pipeline, its own tool execution, its own session storage. When an API changes, when a new model
ships, when a streaming format evolves — that maintenance burden hits every project independently.

Makaio's adapter layer ships conformance-tested integrations for major providers. When an API
changes, one adapter update propagates to every project using the framework. When a new provider
appears, one adapter makes it available everywhere. Shared infrastructure reduces duplicated effort.

The same applies to tools, storage, transport, and session management. The bus is the backbone
that makes sharing possible without coupling: services communicate through typed messages, not
direct imports, so any piece can evolve independently.

## How It Works

Everything in Makaio communicates through a single **typed message bus**. Events, RPC calls,
storage operations, tool invocations — all are bus messages with Zod-validated schemas. The bus is
what makes everything else possible:

- **Adapters** bridge AI providers through a three-layer contract (Adapter > Agent > Connector)
  with shared conformance tests. Each adapter participates in the bus and inherits session
  lifecycle, streaming, and tool execution for free.

- **Sessions** unify conversation history across providers. Start a conversation with one model,
  continue with another. Import sessions from external tools. Fork, branch, and hand off context.

- **Tools** are typed function definitions that register on the bus once and become available to
  every agent. No per-adapter wiring.

- **Extensions** contribute services, tools, CLI commands, HTTP routes, windows, and browser UI.
  The extension model is the primary way to add functionality — from a single tool to a full
  application.

- **Transport** spans process boundaries. The bus protocol is WebSocket + JSON — any application,
  in any language, can participate. VS Code extensions, mobile apps, CI pipelines, external
  tools — connect to the bus and you're a first-class participant.

## Use Your Existing Subscriptions

Most AI tools charge per-seat or per-subscription: Claude Pro, ChatGPT Plus, GitHub Copilot.
You already pay for them. Makaio's adapter model wraps these tools as they are — CLI interfaces,
existing subscriptions, established workflows. You don't need separate API keys and usage-based
billing to get orchestration. Use what you already have.

## Open Framework

The framework is open source under MIT. The [bus](/guides/bus/), [adapters](/guides/creating-adapters/),
[tools](/guides/getting-started/#writing-a-tool), [extensions](/guides/extensions/),
[transport](/guides/transport/), storage, CLI, and desktop host shells — all open, all MIT.

This is deliberate. The bus protocol and extension model are more valuable as shared infrastructure
than as proprietary lock-in. The more tools, adapters, and extensions the community contributes,
the more useful the framework becomes for everyone.

Makaio is developed by Makaio GmbH. The framework is and will remain MIT; commercial products
built on it are in development.

## Where to Start

- [Getting Started](/guides/getting-started/) — Install, run the server, write your first bus interaction, tool, and extension.