Chat Console

Table of Contents

1. What is the Chat Console?
2. Chat Console Requirements
3. Using the Chat Console
4. Models
   1. On-Device
   2. Private Cloud Compute
   3. Claude (Sonnet/Opus)
5. How to Use the Chat Console
6. What Can it Do?
7. Examples

---

PREVIEW

This article covers features only available in beta versions of Drafts.

What is the Chat Console?

iOS Mac

The console is a chat-style interface to use large language models (LLMs) to interact with Drafts. It is not a full-fledged chat client, but think of it as a quick way to do one-off interactions with your content. Each console session is a temporary context to interact with an LLM which has been extended with tools to understand, work with, and modify your drafts.

Chat Console Requirements

The Chat Console is available when you are running:

• Drafts 54 or greater
• Running iOS 27 or macOS 27 or greater
• On a device that supports and has enabled Apple Intelligence

Using the Chat Console

Each chat in the chat console it a individual chat session with a specific Large Language Model. Chats are transient. Drafts does not store a history of chat logs, and each time you start a new one, you are starting a new session with it’s own context and token limits.

Each of these models has it’s own set of capabilities and limitations. Understanding the best model to use for a chat makes a big difference to the quality of the output you will get, so take time to read the “Models” topic below. In addition to each of the model’s own knowledge and capabilities, Drafts provides access to a number of “tools” that allow the models to work with Drafts. Capabilities provided by tools include:

• Creating and updating drafts.
• Querying your drafts library. You can ask find drafts by tags, flagged status, content, etc.
• Work with the editor, opening drafts, getting and changing text and selected text.
• Get or set the system clipboard text.
• Run actions by name.
• Run shortcuts by name.

Models

The chat console supports several different LLMs. Each has its own set of abilities and limitations. Below is a brief summary of some of those to assist you in selecting the best options for your use.

On-Device

The on-device model is based on Apple’s SystemLanguageModel, which ships with the OS.

• Recommended Use Cases
  • Work with a single draft or text selection
  • Summarize draft content, suggest tags
• Pros
  • Private: No data leaves your device when using this model
  • Fast: Tends to be very responsive
  • Works Offline: No network connection required
  • Free and Unlimited: No API keys, no usage limits
• Cons
  • Limited Knowledge: This model is small, and not very useful for general knowledge queries
  • Limited Context Size: A session with the on-device model is limited to 4k tokens. Complex or multi-step prompts will likely result in hitting that limit quickly.

Private Cloud Compute

Apple’s Private Cloud Compute model runs in the Cloud.

• Recommended Use Cases
  • More advanced prompts that require working with multiple drafts or incorporating world knowledge
  • Query and manipulate multiple drafts
  • Create lists of related drafts
  • Run actions across multiple drafts
  • Use world knowledge as input for tasks like “Create a draft with a template for taking Meeting Notes”, “Create a draft with a list of the largest cities in the US”, etc.
• Pros
  • Private(ish): Data does leave your device, but only to Apple’s Private Cloud Compute instances, which promise a high-level of security and do not track and retain any data after completing requests. More details in Apple’s documentation
  • Free (within limits): No accounts or API keys, but you can get rate limited if you use this model a lot. You get a daily limit from Apple, which is higher if you also subscribe to some premium Apple services.
  • More World Knowledge: This model is still not as extensive as frontier models from providers like Anthropic and Google, but has much more general knowledge, so can be useful for tasks that utilize data from outside Drafts.
• Cons
  • Limited Context Size: This model has a 32K token limit per session. Much more generous than the on-device model, but can still be a limiting factor in some cases.

Claude (Sonnet/Opus)

Utilize Anthropic’s Claude models. Currently both Sonnet and Opus are supported.

• Recommended Use Cases
  • Advanced prompts that require working with multiple drafts or incorporating world knowledge
  • Rewriting and proofing text
  • Query and manipulate multiple drafts
  • Create lists of related drafts
  • Run actions across multiple drafts
  • Use world knowledge as input for tasks like “Create a draft with an outline for a trip to Los Angeles”, “Create a draft with a list of the largest cities in the US”, etc.
• Pros
  • Powerful: These options use some of the most advanced Claude models and all the power and capabilities they offer.
• Cons
  • Not Free: Usage requires a Claude API Key and usage will be counted against your Anthropic account and usage limits.

The first time you use one of the Claude Models, you will be prompted to enter an API. You can generate API keys at the Claude Console.

How to Use the Chat Console

What Can it Do?

Drafts makes available to each of the models a series of “Tools” that enable functions in the app. Core features that are available to the LLMs include:

• Create and update drafts
  • Set the text, append, and prepend to drafts
  • Assign or remove tags, set flagged status
• Query drafts
  • Find drafts by tags, text searches, flagged status, etc.
• Open Workspaces
• Run Actions by name
• Work with the editor
  • Open drafts
  • Get and set selected text and selection ranges
• Get or set the system clipboard

Examples

The chat console is open-ended, but here are a few ideas of the types of things it can be good at.

• Simple Prompts (Best with On-Device)
  • “Create a draft with the content of the clipboard”
  • “Suggest and assign tags for the current draft”
• Advanced Prompts (Best with PCC/Claude)
  • “Find drafts assign the tag ‘project-a’ and run the “Save to Project Folder” action on each”
  • “Create and open a new draft with a summary of the my flagged drafts”
  • “Create a draft with a sample one-day trip itinerary for a visit to Seattle”
  • “Reformat the selected text as a Markdown list, separating using commas”
  • “Rewrite the selected text in a more formal tone”

---