Contents

Guides

Why Does Ritemark Use Terminal AI Agents?

•7 min read

Why Does Ritemark Use Terminal AI Agents?

When you use the AI sidebar in Ritemark, the app connects to a local agent runtime: Claude Code, Codex, or OpenCode. Claude Code and Codex are official tools from Anthropic and OpenAI; OpenCode is an open-source agent that can connect to several model providers. This article explains why Ritemark uses these runtimes and what that means for your privacy.

What Are Terminal AI Agents?

Terminal AI agents are programs that run on your computer and work with your files directly. Unlike browser-based chatbots where you copy-paste text, terminal agents:

  • Read your files from disk, so you never have to copy-paste content

  • Read, propose, or make edits according to the permission mode and approvals you choose

  • Run the agent process locally on your machine while cloud-connected models still process requests on provider servers

  • Understand context by reading multiple files in your project at once

Claude Code (by Anthropic), Codex (by OpenAI), and OpenCode are the runtimes Ritemark supports. They follow the same broad pattern: you give an instruction in plain language, the agent reads permitted context, uses tools, and returns an answer or file changes. Whether each action needs approval depends on the mode you select.

Why Does Ritemark Use External CLIs?

They Follow Their Own Upstream Release Cycles

Anthropic maintains Claude Code, OpenAI maintains Codex, and the OpenCode project maintains its open-source agent. Their releases can add models, capabilities, and fixes independently of Ritemark.

If Ritemark built its own AI agent from scratch, every model update would require engineering work on our side. By using the official CLIs, you get access to new AI models on the day they launch, without waiting for a Ritemark update.

Your Files Stay Local, While AI Context Can Be Sent

The agent process reads files from your disk, and Ritemark does not upload your folder to a Productory document server. When you invoke cloud-connected AI, however, your prompt and relevant context can be sent to the selected provider. This can include the active file, selected text, attachments, shared browser context, and tool results.

Ritemark does not proxy AI requests through a Productory server. The selected runtime connects to its provider using the account or API key configured for that runtime.

You Control Your Own Account and Billing

You sign in with your own provider account or add your own API key. Credentials are handled locally by Ritemark's secret storage and the installed runtime; Productory does not receive the key or process provider payments on your behalf.

This means you can:

  • Use a supported Claude or ChatGPT subscription where the runtime permits it

  • Set your own spending limits in the provider's dashboard

  • Switch among configured runtimes and providers from Ritemark's model picker

They Work With Any File, Not Just the Open Document

Terminal agents can read and edit any file in your project directory. That makes them useful for batch operations like translating an entire folder of documents, restructuring a documentation set, or applying consistent formatting across many files.

A built-in AI that only sees the open document would be limited to single-file edits.

Why Not Build AI Directly Into Ritemark?

Building and maintaining an AI agent is a large undertaking. Here is what it would mean:

Slower updates. When Anthropic releases a new Claude model, Claude Code supports it immediately. A custom-built agent would need engineering time to add each new model, test it, and ship an update.

More bugs. AI agents handle complex tasks: file reading, diff generation, permission management, authentication, rate limiting. The official CLIs have dedicated teams testing these features. A smaller team would produce a less reliable tool.

Higher cost for users. Running AI through Ritemark's own servers would mean Ritemark pays for compute and passes costs to you, with markup. By using the official CLIs, you pay the AI provider directly at their published rates, or use your existing subscription.

Security trade-offs. Handling API keys and user authentication is sensitive work. The official CLIs are built and audited by the AI companies themselves. Ritemark avoids this responsibility entirely by letting the CLI handle auth.

What Does Ritemark Add on Top?

Ritemark is not just a pass-through. It adds a layer of convenience and integration that makes terminal agents more accessible:

  • Visual sidebar UI -- chat with AI agents without touching the terminal

  • Runtime and model switching -- pick Claude Code, Codex, or OpenCode and an available model from the composer

  • Document context -- Ritemark passes your open document and project context to the agent automatically

  • Onboarding wizard -- detects missing dependencies (Git, Node.js, CLI tools) and guides you through installation step by step

  • Settings integration -- see agent status, authentication method, and CLI version at a glance

You get the power of terminal agents with the ease of a graphical interface.

How Does Privacy Work?

Here is the data flow when you use an AI agent in Ritemark:

  1. You type a request in Ritemark's AI sidebar

  2. Ritemark passes the request and relevant file content to the CLI agent running on your machine

  3. The agent sends the request to the selected provider or service using your account credentials

  4. The AI provider processes the request and returns a response

  5. The CLI agent shows you the proposed changes

  6. Ritemark shows progress, results, and any approval request required by the selected mode

What Productory receives from the AI request: Ritemark does not route the request or your folder through a Productory server. The desktop app keeps local chat history and files on your device.

What the AI provider sees: The prompt and context sent by the runtime, processed under the provider terms and account settings you use.

What stays on your machine: Your files, project, git history, and local chat storage remain local as stored data. File content can still be transmitted as AI context when the agent reads or sends it for a task.

Separate Ritemark analytics: When enabled, the app sends anonymous product-usage events to PostHog. These events do not include prompts or file contents; written feedback text is sent only when you explicitly submit it. See AI Information in Ritemark for the full data-flow description.

Frequently Asked Questions

Do I have to use the terminal to use AI in Ritemark?

No. Ritemark's AI sidebar provides a visual interface for terminal agents. Select an agent from the dropdown, type your request, and the sidebar handles the rest. The terminal is available as an alternative for advanced users.

Can I use AI in Ritemark without installing a CLI tool?

The sidebar runtimes require their respective local agent components. Ritemark's setup flow detects what is installed and guides you through the supported authentication options.

Is my data safe when using AI agents?

Ritemark does not send your folder through a Productory AI server, but the selected runtime can send prompts and relevant file or tool context to its provider. Check AI Information in Ritemark and the policies for the provider and account you use.

Which agent should I choose -- Claude Code or Codex?

Choose based on the provider account, models, and workflow you want to use:

  • Have a Claude Max subscription? Use Claude Code.

  • Have a ChatGPT Plus or Pro subscription? Use Codex.

  • Have both? Try each and pick your preferred experience.

  • Want to use your own Google, OpenAI, Anthropic, or OpenRouter API key? Configure OpenCode.

You can switch agents at any time from the sidebar dropdown.

Will Ritemark support more agents in the future?

Ritemark currently supports Claude Code, Codex, and OpenCode through one shared runtime interface. Additional runtimes require a deliberate product and engineering decision.

What happens if I do not install any AI agent?

Ritemark works perfectly as a markdown editor without any AI agent installed. AI features are optional. You can write, edit, and manage documents without them.

Back to support