Chat Interface

BB's chat interface is your primary workspace for AI collaboration. Learn how to choose personas, follow agents as they work, answer BB's questions, and work alongside others in the same conversation.

Starting Your First Collaboration

When you create a new collaboration, BB presents a helpful starting interface with categorized task examples to inspire your objectives:

New Collaboration Interface
BB's starting interface with categorized task examples and objective guidance

Task Categories

  • 🔍 Research Tasks - Analysis, literature reviews, data exploration
  • ✏️ Content Tasks - Writing, editing, content strategy
  • 📊 Analysis Tasks - Data visualization, metrics analysis
  • đź’» Development Tasks - Coding, refactoring, system design

đź’ˇ Pro Tip

Focus on what you want to achieve, not implementation details. BB works best when you describe your objectives clearly.

Collaboration Settings & Personas

BB works in three roles: the Orchestrator plans and answers you, Agents take on tasks it hands out, and Admin handles housekeeping such as titles and summaries. Each role runs with a persona: a named setup with its LLM and model settings, guidelines and tool sets. Your project's defaults choose a persona for each role, so most of the time there's nothing to change. When one conversation needs something different, such as a more capable LLM for a hard problem, you can change it here without touching your project settings.

Open Collaboration Settings with the gear icon at the left of the message input's toolbar. It has two parts: Personas, and Chat Settings, where Save Memories In chooses where memories from this chat are saved.

Collaboration Settings dialog
Collaboration Settings, opened from the gear in the message toolbar

Choosing a Persona for Each Role

  • One picker per role: Orchestrator, Agent and Admin. The persona the conversation uses now reads "name (in use)".
  • "Default: name": switches the role to whatever your project's current default persona is.
  • Badges show where each persona comes from: Built-in, Workspace, Project or This conversation. Below the picker, a caption shows the LLM it uses ("Model: role default" when it uses the role's usual LLM).
  • When changes apply: they are saved with the conversation when you send your next message, for everyone in it.

Why a conversation keeps its persona

A conversation keeps the persona it started with, even if someone later changes the project's or workspace's defaults. That way a long-running piece of work doesn't suddenly change LLM or tools halfway through. To move a conversation onto the new default, pick "Default: name". More about personas in conversations

Customising a Persona for One Conversation

Customise for this conversation makes a copy of the role's persona, named "name (this conversation)", that only this conversation uses. You can change its LLM and model settings: temperature, maximum response length, extended thinking, prompt caching and when context management starts. Tools, guidelines and limits stay as the persona sets them; to change those, edit the persona itself. Stop customising returns the role to its default persona.

Model selection dropdown with a search box and a list of models
Choosing an LLM for a customised persona

Switching LLMs can cost more

Once a conversation has run, BB shows a notice when a change may use more credits. For the Orchestrator and Agent roles it appears whenever the LLM changes: the next message (or the next new agent) can't reuse the context already saved for the old LLM, so it may cost more once. For Admin it appears only when the new LLM is priced higher.

Deprecated and retired LLMs

An LLM that is being phased out carries a Deprecated badge in the picker. Hover over it to see when it retires and what replaces it, and BB shows a notice naming the retirement date and the recommended replacement.

If a conversation or persona still uses an LLM after it has retired, BB uses the replacement automatically and notes it in the chat ("Model has been retired; using Replacement instead."). Claude Sonnet 4.5 through Anthropic's own service has retired in this way.

Agents in the Conversation

For bigger requests, the Orchestrator hands parts of the work to agents and keeps going while they run. You can watch them work, steer them, or stop one, without stopping everything else.

Agent groups show when the message filter is All or Tools. The default, Conversation, hides them (see Message Filtering).

Three agents working in parallel: Marketing plan vs sales, Menu vs survey and Wholesale price change notes, each Running with its persona, who started it, elapsed time, tool calls and latest progress, and Message and Cancel buttons
Agents working in parallel, each in its own group
  • Agent groups: each agent shows its task, persona, status (such as Queued, Running, Waiting or Completed), how long it has been working and its latest progress. Agents started by other agents appear in a group nested inside.
  • Message: send an agent a note. It reads your message after its current step, and everyone in the conversation sees it as yours.
  • Cancel: stop one agent. The answer and the other agents carry on. Cancelling someone else's agent asks you to confirm first.
  • Status bar: describes in plain words what BB is doing, such as "Starting 3 agents", with a line for each running agent ("Thinking...", "Waiting for an answer", "Waiting for its agents").
  • "As account, started by name": in shared team workspaces, an agent may run as an agent account rather than as the person who started it. The label on its group and report says which.

See Orchestrator & Agents for every status and what it means.

Questions from BB

Sometimes BB, or one of its agents, needs a decision from you before it can go on well. Rather than guessing, it asks. The question appears as a card in the conversation, marked "Waiting for an answer."

Question from Claude asking which wholesale accounts should get the new January prices, with a Your answer field, Send answer and Decline buttons and Waiting for an answer.
BB asks a question and waits for an answer
  • Type your reply and choose Send answer, or Decline to let BB carry on without one.
  • Anyone who can write in the conversation can answer, not only the person who sent the message.
  • Answered questions stay in the history marked Answered, Declined or Withdrawn.

Questions are on by default. Approval requests, where BB asks before it changes something, work in a similar way but are off unless you turn them on. See Questions & Approvals.

Run My Agents as Me

In shared team workspaces, agents may run as an agent account with its own access. If you'd rather your agents use your own access for a while, turn on Run my agents as me in the message toolbar ("As me" on narrow screens). It only appears when your agents would otherwise run as an account. While it's on, agents started for your messages, and the agents they start, run as you. It's off by default and applies to this conversation until you reload the page. More about Run my agents as me

Signing In to Connected Services

Some connected services (MCP servers) need each person to sign in with their own account. When one of those is available, a notice above the message input reads "Some tools need you to sign in".

  • When BB needs such a service, a Sign in to server card appears with Sign in and Not now. Once you've signed in, BB continues where it left off.
  • Others in the conversation see that BB is waiting for you to sign in.

See MCP Servers for setting up connected services.

Working with Others in a Conversation

In a shared workspace, several people can write in the same conversation. Messages sent while BB is busy wait in a queue, the status bar says whose message BB is answering, and the Stop button splits into "Drop my message" and "Halt all messages" so you can stop your own request without stopping everyone else's. See Working together in a collaboration.

Collaboration Management

BB provides powerful tools to organize and navigate your collaborations efficiently:

Collaboration Selector with Search
Search collaborations and add favorites for quick access

Organization Features

  • Search: Find collaborations by title or content
  • Favorites: Star important collaborations for quick access
  • Auto-titles: BB generates descriptive titles from your objectives
  • Token tracking: See usage and turn counts at a glance
  • Timestamps: Track when collaborations were last active

Best Practices:

  • Create focused collaborations for specific objectives
  • Use favorites for ongoing projects
  • Start fresh collaborations to avoid token buildup

Understanding Message History

Chat Message History with Expandable Entries
Message history showing different entry types with expand/collapse functionality

Entry Types & Colors

BB's message history includes several types of entries, each with distinct colors for easy identification:

User Messages (Blue): Your prompts and instructions to BB
Orchestrator (Purple): BB's high-level planning and task coordination
Assistant (Green): BB's detailed responses, analysis, and recommendations
Answers (Dark Green): Final answers and summaries from BB
Tool Operations (Yellow/Orange): Tool inputs showing what BB is doing (file operations, searches, etc.)
Tool Results (Orange): Output and results from tool operations
Agent Tasks (Dark Blue): Work handed to agents, with their progress and reports (see Agents in the Conversation)
Auxiliary (Gray): System messages, metadata, and supporting information

Note: You can control which entry types are visible using the message filtering feature.

🔍 Expandable Entries - Explore Everything!

Every entry in the collaboration history can be expanded or collapsed. While user prompts and AI responses are shown expanded by default, all other entries (tool operations, system messages, etc.) start collapsed to keep the interface clean.

What You Can Expand:

  • Tool input parameters and settings
  • Complete tool output and results
  • System operation details
  • File modification summaries
  • Error messages and diagnostics

How to Expand:

  • Click the entry header to expand/collapse
  • Space key when entry is focused
  • Arrow icons show expand/collapse state

đź’ˇ Pro Tip: Expanding tool entries helps you understand exactly what BB is doing and learn how different operations work. It's a great way to build your understanding of BB's capabilities!

Message Filtering & Organization

Keep your collaboration view clean and focused with BB's message filtering system. Control which entry types are visible with quick-access filter presets or create custom views tailored to your workflow.

Message Filtering Interface
Filter bar with presets and entry count summary

Filter Presets

Access filter presets from the header above your collaboration. The entry count summary shows how many entries are visible and which types are hidden.

All

Shows every entry type in the collaboration—user messages, BB responses, tool operations, agent tasks, and system messages.

Best for: Debugging, understanding BB's complete workflow, learning how tools work

Conversation (Default)

Focuses on the dialogue—your messages, BB's responses, and final answers. Hides technical details like tool operations and agent tasks.

Best for: Clean reading experience, focusing on outcomes, reviewing collaboration flow

Main Flow

Shows user messages, orchestrator planning, assistant responses, and answers. Provides insight into BB's decision-making without tool-level detail.

Best for: Understanding BB's reasoning, following the high-level workflow

Tools

Highlights tool operations, results, and agent tasks. Shows what actions BB is taking behind the scenes.

Best for: Monitoring file changes, tracking tool usage, verifying operations

Custom Filtering

Click "+ More" to access granular filtering options. Select exactly which entry types you want to see. Your filter preferences save automatically per collaboration.

Available Entry Types:

See the Message History section above for detailed descriptions of each entry type:

  • User Messages
  • Orchestrator
  • Assistant
  • Answers
  • Tool Use
  • Tool Results
  • Agent Tasks
  • Auxiliary
  • Errors (shown in every preset)

Smart Features

  • Filters save per collaboration
  • Entry counts update in real-time
  • Hidden entries show as "(hidden)" in counts
  • Collapse states reset when filter changes

Keyboard Shortcuts

Use keyboard shortcuts for quick filtering. See the Keyboard Shortcuts section below for the complete list.

Editor Modes: Visual and Markdown

The message input has two editor modes. Switch between them with the Visual / MD toggle in the toolbar at the bottom of the message input. BB remembers your choice in that browser, and your message carries over when you switch.

Message input in Visual mode with a selected word in bold, the formatting toolbar above it (B, I, code, H1, H2, H3, bullet and numbered list buttons), a researchers.tsx [local] file chip, and the Visual / MD toggle in the input toolbar
Visual mode: select text to format it, and linked files show as chips. The Visual / MD toggle is in the toolbar below your message.

Visual (default)

A formatted editor: you see bold text, headings, and lists as they'll appear, rather than the Markdown behind them.

  • Select text to open a formatting toolbar: bold, italic, inline code, headings 1 to 3, bullet and numbered lists
  • Type @ to add files and other resources
  • Linked resources and attachments appear as chips you can see at a glance
  • Paste or drag files into the editor to attach them
  • Pasted Markdown is converted to formatting

Best for: everyday messages, and anyone who'd rather not write Markdown

Markdown (MD)

A plain-text editor: you write Markdown directly and see exactly what BB will receive.

  • Type / at the start of a path, or press Tab, to add files
  • Selected files are inserted as paths in backticks, with the data source added when you have more than one
  • Paste an image to attach it
  • Chips from Visual mode show as [[...]] references, and become chips again when you switch back

Best for: precise control over Markdown, and typing file paths quickly

Markdown mode message input reading 'Look at [[local:docs/quick-start/researchers.tsx]]', with MD selected in the toolbar
Markdown mode: a linked file shows as a [[data source:path]] reference instead of a chip

Both modes: Enter adds a new line and Cmd/Ctrl + Enter sends.

File Suggestions & Autocomplete

BB provides intelligent file suggestions across all your data sources with advanced autocomplete featuring multi-datasource support and wildcard patterns. How you open suggestions depends on your editor mode:

Visual mode message input reading 'Finish the December newsletter draft in @newsletter', with the suggestion newsletter-draft-december.md showing its folder (marketing) and data source [Fernhill Coffee Roasters]
Visual mode: type @ and part of a file name to see matching files
Markdown mode message input containing quick-start, with MD selected in the toolbar and a suggestion list of matching files such as researchers.tsx and writers.tsx, each showing its folder and [local] data source
Markdown mode: press Tab to see matching files and folders
Visual mode message input asking to finish the December newsletter, with file chips for newsletter-draft-december.md and brand-voice.md from the Fernhill Coffee Roasters data source
In Visual mode, the file you select appears as a chip in your message

Trigger Methods

  • Visual mode: Type @, then at least two characters of a file name. The file you pick is added as a chip.
  • Markdown mode: Type / at the start of a path, or press Tab anywhere in your message. The file's path is inserted as text.
  • Data source context: Each suggestion shows which data source it comes from

Advanced Patterns

  • *.ts - All TypeScript files
  • docs/**/* - All files in docs subdirectories
  • src/**/test* - Test files at any depth in src

Navigation & Selection:

Keyboard: ↑/↓ navigate • Enter selects • Escape closes
Best Practice: List files on separate lines for clarity

Prompt History & Reuse

Efficient Prompt Reuse

Access your recent prompts through the clock icon in the message input's toolbar. Perfect for iterating on ideas or reusing common instructions.

Recent prompts dropdown above the message input, listing two earlier prompts with how long ago they were sent and a + button to reuse each one
Quick access to recent prompts for reuse and iteration

Key Features

  • Recent prompts: Last several messages you've sent
  • Quick selection: Click to reuse or modify previous prompts
  • Time-saving: Avoid retyping similar requests

Use Cases

Great for refining prompts, trying variations, or building on previous successful interactions.

Model Information & Context Monitoring

Click the "Model Info" button in the header to see detailed information about your current model configuration and context usage:

Detailed Model Information Panel
Comprehensive model stats including context usage, caching impact, and configuration details

Context Tracking

  • Current token usage vs. limit
  • Percentage of context window used
  • Real-time usage monitoring

Caching Benefits

  • Prompt caching status
  • Token savings from cache
  • Cost reduction impact

Configuration

  • Current model details
  • Temperature and settings
  • Extended thinking status

Keyboard Shortcuts & Productivity Tips

Message Controls

Send messageCmd/Ctrl + Enter
New line in messageEnter
File suggestions (Visual mode)@
File suggestions (Markdown mode)
/orTab
Stop BB's answer (while it's working; in a shared conversation, "Drop my message")
Cmd/Ctrl + .orEsc

Message Navigation

Expand/collapse entrySpace
Navigate entries
↑↓
Navigate file suggestions
↑↓

Message Filtering

Filter Controls

Toggle filter dropdownf
Close filter dropdownEsc
Toggle checkbox (when focused)
SpaceorEnter

Quick Presets (when dropdown open)

Select "All" preset1
Select "Collaboration" preset2
Select "Main Flow" preset3
Select "Tools" preset4

Token Management & Cost Optimization

Context Usage Monitoring

The progress bar below the message input shows how much of the model's context window is being used. Monitor this to understand when collaborations are approaching context limits.

Context progress bar below the message input, with a 44K token label on the left and 4.0% of context on the right
Real-time context window usage, shown in tokens and as a percentage of the context window

Color Indicators

Green = Normal usage (under 50%)
Yellow = Moderate usage (50%+ context used)
Red = High usage (90%+ context used)

Tiered Pricing Models

For models with tiered pricing, tick marks show where price tiers change, and the label color shifts toward red as you progress through higher cost tiers.

Token Management Tips

  • Monitor the progress bar below the message input
  • Keep collaborations focused on single objectives
  • Start fresh collaborations to reset context
  • Remove files when no longer needed

Cost Optimization

  • Enable prompt caching for repeated context
  • Use collaboration summaries for long sessions
  • Choose appropriate models for different tasks
  • Leverage BB's auto top-up for budget control

Pro Tip: The Model Info panel shows detailed caching benefits and context usage. Use it to understand your token efficiency and optimize your workflow.

Getting Started with Advanced Features

Quick Start Checklist

Essential Setup:

  • Check which persona each role uses in Collaboration Settings
  • Enable prompt caching to reduce costs
  • Set up auto top-up for uninterrupted usage
  • Explore the expandable message entries

Productivity Tips:

  • Add files with @ (Visual mode) or Tab (Markdown mode)
  • Star important collaborations as favorites
  • Use message filtering to focus on dialogue or tools
  • Reuse prompts from the history dropdown
  • Monitor token usage with the progress bar

Next Steps

Last updated: November 16, 2026