# moxxy

> Your local AI agent. Simple to start. Ready to govern.

Moxxy is an open-source, local-first agent for developers. It works in the
current project with a model account the developer chooses, persists runs
locally, and routes consequential tool calls through approval and policy.
Moxxy is currently a developer alpha. It requires Node.js 20.10 or newer and is
MIT licensed.

## Personal golden path

```sh
npm install -g @moxxy/cli
cd your-project
moxxy onboard
moxxy
```

Default onboarding asks the developer to choose one model connection and
authenticate. It selects the recommended model, runtime mode, and local memory
index automatically. No config file, channel, service, compactor, cache
strategy, embedder, or isolator decision is required.

Supported connection types include Anthropic and OpenAI API keys, ChatGPT
OAuth, and an existing Claude Code subscription session.

## Use

- A workspace is the project moxxy can work in.
- A run is a conversation and its activity.
- Read-only work stays light; writes and commands ask for approval.
- Sessions use an append-only event log and can be resumed.
- The CLI and desktop share the same runtime.

## Extend

Extensions are optional capabilities added after the core works. Use
`moxxy extensions` to manage them. Skills are prompt-only Markdown. Code
extensions use the typed, zero-runtime-dependency `@moxxy/sdk` plugin contract.

Advanced authors may contribute tools, model providers, channels, modes,
compactors, cache strategies, embedders, isolators, and audit sinks. Those are
author concepts, not first-run product concepts.

## Govern

Governed moxxy applies an organization profile to the same local developer
experience. Profiles can constrain model connections, extensions, tools,
network egress, and audit export. Identity is attached at the channel boundary,
and bounded redacted receipts explain what happened.

Governed workstations are a design-partner preview. Moxxy does not yet claim a
fleet control plane, compliance certification, or production SLA. Hash-chained
local audit data is tamper-evident, not tamper-proof; isolation is opt-in.

## Developer alpha feedback

Useful feedback includes OS, architecture, Node version, install method, model
connection, time to first useful result, first attempted task, unclear words or
approvals, and whether resume worked. Never include secrets.

- Bugs: https://github.com/moxxy-ai/moxxy/issues
- Product feedback: https://github.com/moxxy-ai/moxxy/discussions

## Links

- Site: https://moxxy.ai
- Documentation: https://docs.moxxy.ai
- Source: https://github.com/moxxy-ai/moxxy
- npm: https://www.npmjs.com/package/@moxxy/cli