Chat

Overview

Chat is the primary way you interact with Thoughtful's AI assistant. It combines the power of Claude with deep access to your workspace context, enabling you to ask questions, manage pages, search across your content, and collaborate with team members in real time.

Chat Threads

When you open Thoughtful, you start in a chat thread. Threads are persistent conversations that save your entire history with the AI. You can:

• Start new threads by clicking the compose button in the header
• Switch between threads using the thread menu dropdown
• Rename threads to help you find them later (or let the AI auto-generate a title based on your first message)
• Delete threads when you no longer need them
• Pin threads to keep important conversations at the top of your list

Private by Default

Every chat thread you create is private to you. Only you can see it in your thread list, and only you can read or send messages in it.

Multi-User Threads

You can invite team members to join your chat threads for collaborative conversations with the AI. To add someone:

• Click the participants button in the chat header (shows avatar icons)
• Select team members from the dropdown
• Or simply @mention a team member in your message - Thoughtful will ask if you want to invite them

When team members join a thread:

• Everyone sees the full conversation history
• Messages show who sent them with author names
• Real-time typing indicators show when others are composing
• Online presence dots show who is currently viewing the thread

The thread owner (the person who created it) can add or remove participants. Participants can choose to leave a thread at any time.

Real-Time Streaming

When you send a message, the AI's response streams to your screen in real time - you see each word as it's generated, not just the final result. This makes long responses feel more interactive and lets you stop generation early if needed.

Multi-User Streaming

In shared threads, streaming works for everyone simultaneously:

• When one person sends a message, all participants see the AI's response stream in real time
• A "lock" indicator shows who triggered the current response
• Your messages are queued if someone else is currently streaming, then sent automatically when they finish

This coordination prevents confusion and ensures everyone sees the same conversation state.

Context Window Indicator

Near the chat header, you'll see a circular progress indicator showing how much of the AI's context window is being used. This matters because AI models have a limited memory for each conversation.

Click the indicator to see a detailed breakdown:

• System prompt: Base instructions that define how the AI behaves
• Tool definitions: Descriptions of all the actions the AI can take
• Messages: Your conversation history

As conversations grow longer, you'll use more of the context window. When you approach the limit (indicated by orange/red coloring), consider starting a new thread to give the AI fresh context.

Sub-Agents: Preserving Context

For complex, long-running tasks that might consume your entire context window, the AI can spawn a sub-agent. This is a separate AI instance that:

• Gets its own fresh context window
• Has access to all the same tools as the main conversation
• Works independently on a focused task
• Reports its results back when finished

You'll see sub-agent tasks appear as expandable cards in the conversation. You can watch their progress in real time, seeing their thinking process, tool calls, and results. When a sub-agent completes, its summary is added to your conversation without using up your main context.

This is particularly useful for tasks like:

• Exploring large codebases
• Generating detailed content
• Running multiple related operations

@Mentioning Pages and People

Type @ in the chat input to mention pages or team members:

Pages

• Type @ followed by a page name to search your workspace
• Select a page to include it as context for the AI
• The AI will read the page's content to answer your questions

People

• Type @ followed by a team member's name
• If they're not already in the thread, you'll be asked to invite them
• The AI recognizes user mentions and can factor them into responses

Mentions appear as styled chips in your message, making it clear what pages and people are referenced.

Workspace Context

The AI has deep access to your workspace through a comprehensive set of tools:

Page Operations

• List and search pages
• Create, update, and organize pages
• Edit page descriptions with find/replace
• Add comments and suggestions
• Manage page properties and status

Search and Discovery

• Full-text search across all content
• Semantic similarity search
• Search chat history from other threads
• Find pages by owner, status, or relationships

Integrations

When connected, the AI can also access:

• GitHub: Browse repos, PRs, commits, and code
• Slack: Read threads and channel activity
• Google Drive: Search and read documents and spreadsheets
• Readwise: Search highlights and saved articles
• Web: Search the internet and fetch content from URLs
• MCP servers: Connect custom tools via the Model Context Protocol

Skills and AGENTS.md

You can customize how the AI behaves in your workspace:

• AGENTS.md: Create a root-level page named "AGENTS.md" with custom instructions that the AI will follow
• Skills database: Create reusable prompts and workflows in a "Skills" database that the AI can reference

Chat History

All your conversations are automatically saved and searchable:

• Messages persist even if you close the browser
• Search across all your threads to find past discussions
• The AI can search your chat history to find relevant past conversations
• Thread activity timestamps help you find recent discussions

Threading Model

Each thread maintains:

• Complete message history with timestamps
• Author attribution for multi-user threads
• Read/unread state per participant
• Last activity timestamp for sorting

Attachments

You can attach files to your messages:

• Images: Paste or drag-and-drop screenshots and photos
• Documents: Upload PDFs, spreadsheets, and text files
• Links: Paste URLs to have their content automatically fetched

Large files are processed asynchronously - the AI gains access to their content through dedicated reading tools rather than embedding everything in the conversation.

Voice Input

Click the microphone button to dictate your message:

• Speak naturally into your microphone
• Your speech is transcribed automatically
• Review and edit the text before sending
• Send as normal or continue typing

This is especially useful on mobile or when you want to capture thoughts quickly.

Keyboard Shortcuts

Tips for Effective Chat

• Be specific: Clear, detailed requests get better results
• Use @mentions: Reference specific pages to give the AI focused context
• Watch the context: Start fresh threads for new topics to avoid hitting limits
• Let sub-agents help: Complex tasks often work better when the AI can spawn focused workers
• Collaborate: Invite team members to threads where you're working on shared problems