@moxxy/mode-deep-research

Deep-research loop strategy for moxxy: decompose the user question into parallel sub-queries, fan out via subagents, synthesize a cited writeup.


Keywords
moxxy, agent, mode, research, fan-out
License
MIT
Install
npm install @moxxy/mode-deep-research@0.38.0

Documentation

moxxy

moxxy

Your local AI agent. Simple to start. Ready to govern.
Work in a project with your preferred model, explicit approvals, and optional extensions.

CI status Node.js 20.10 or newer MIT License Developer alpha

Start · Extend · Govern · Documentation · Contributing

What moxxy is

Moxxy is an open-source agent for software work. It runs locally, understands the current workspace, and asks before consequential tool calls. Connect an API key, ChatGPT account, or Claude Code subscription and start from the terminal or desktop app.

The product has two paths built on the same runtime:

  • Personal use: a developer connects a model and works in a project with sensible defaults.
  • Governed use: an organization supplies policy, approved model connections and extensions, and inspectable activity receipts.

The modular TypeScript runtime remains available to extension authors, but its internal modes, compactors, cache strategies, embedders, and isolators are not part of first-run setup. See the product contract for the boundary.

Start in a project

Requirements: Node.js 20.10 or newer and one supported model account.

npm install -g @moxxy/cli
cd your-project
moxxy

On first start, Moxxy asks you to choose a model connection and sign in or add an API key, then opens the project TUI. Safe workspace reads proceed directly; commands, changes, and external access are described before approval. No config file is required.

Ask one question without opening the interactive UI:

moxxy -p "explain how authentication works in this repository"

Use moxxy help advanced when you want channels, services, automation, or operator commands. See Getting started for all supported authentication paths.

What you get

Local workspace context The agent works in the project you launch it from and persists runs locally.
Clear approvals Commands, changes, and external access are presented by target and impact.
Model choice Use first-party API providers or subscription-backed ChatGPT and Claude Code connections.
Long-running work Continue saved runs with bounded context and optional long-term memory.
CLI and desktop Use the same runtime from a keyboard-first terminal or native desktop workspace.
Open extension surface Add capabilities without forking the core or locking into one model vendor.

Extend when you need to

moxxy extensions finds, installs, enables, and removes optional capabilities. Skills are Markdown; code extensions use the typed @moxxy/sdk contract.

moxxy extensions list
moxxy extensions search browser
moxxy extensions install browser

Advanced authors can contribute tools, model connections, channels, modes, compactors, cache strategies, embedders, and isolators. Start with the developer guide.

Governed workstations

The governed path applies an organization profile to the same local developer experience. Profiles can constrain model connections, extensions, tools, network egress, and audit export while verified receipts explain what happened in a run.

Governed use is a developer-alpha preview for design partners; moxxy does not yet claim a fleet control plane, compliance certification, or production SLA. Read Deploying in an organisation, the threat model, and what leaves the machine.

Developer alpha

The alpha is for developers willing to test installation, authentication, first useful work, approvals, and resume behavior. Limitations and the feedback format are documented in Developer alpha.

Please report reproducible problems through GitHub Issues and product feedback through GitHub Discussions.

Documentation

License

Moxxy is available under the MIT License.