How Nock works

Nock connects four layers: voice, a model that understands the request, tools that do the work, and other agents you choose to involve. Configure the speech and AI connections independently.

You speak or type
       │
       ▼
Voice: local recognition → optional cloud speech
       │
       ▼
Brain: your connected AI or Nock's included AI
       │
       ├── Hands: local V3Code engine, workers, files, browser, board
       │
       └── Connections: coding chats | Team messages | cloud bots

Voice

Recognition starts on your computer. Nock prepares a speech model in the background, and compatible Macs can also use Apple’s on-device recognizer. Automatic can use an eligible cloud route when local recognition cannot keep up or is unavailable; Never send my voice off this computer prevents that fallback. Light cleanup and replacements also run locally. Spoken replies can use an on-device voice, independently of the model that wrote them. See Local dictation, Dictation cleanup, and On-device voice.

Brain

The model interprets your request, chooses tools, and writes the answer. Use Nock’s included AI, connect your own provider key, or let your signed-in Codex think on your ChatGPT plan. The model receives the request and relevant context; local recognition does not make an agent conversation offline. Nock’s Privacy page summarizes your current settings and general service handling. Models and providers explains included routes and Bring your own AI explains personal connections.

Hands

The bundled V3Code Terminal engine works in a project folder on your computer: local indexing, code search, language tools where available, edits, commands, browser work, and background workers. A worker can continue while Nock answers your next question. Tasks and worker conversations let you inspect progress, steer work, review files, and stop a job. Glass Canvas is a visible board the agent can read and draw on. Start with Coding agents, Projects, and Background tasks.

Delegates

Nock can work with your existing Claude Code, Codex, and supported coding chats. The Chat Manager helps find them, send instructions, and answer requests for input. Team supplies message destinations for selected chats shared with people, computers, and outside agents; sharing does not start a job or send the existing transcript. Incoming messages wait for delivery approval unless that chat has automatic delivery enabled. A NockBot instead executes work on its own cloud computer, with its own workspace and fair-use policy. Chat Manager, Team, and NockBot explain each connection.

Follow one request

For “have Codex review this project,” local speech becomes text. Nock identifies the request and project and starts the supported coding workflow using your Codex connection. Depending on the connected client and request, work appears in a coding chat or a background task. Open that conversation or task to review progress and the result. A local speech model, a provider subscription, a project folder, and a Team or cloud connection are separate choices—you can use one without enabling all the others.