Getting Started with PebbleChat

This page walks through the PebbleChat interface element by element, then takes you through your first real conversation. If you know your way around and just want to start chatting, skip to Your first conversation.

Opening PebbleChat

Click PebbleChat at the top of the left sidebar. That’s it — there’s no separate “open chat” step. You land directly in the chat interface, ready to type.

Interface tour

The PebbleChat screen has four main areas:

1. Conversation sidebar (left)

A collapsible panel listing every conversation you’ve had in this workspace. It opens and closes with a toggle button — closed gives you more room for the chat itself, open is better when you’re jumping between threads.

The sidebar contains:

• New Chat button at the top
• Search Chats box
• Date groups (Today, Last 30 days, Older)
• Each past conversation listed by its auto-generated title — click to resume
• A spinner icon next to any conversation still running in the background

2. Main chat area (centre)

Empty when you arrive (the “landing” view), populated as soon as you send your first message.

On the landing view you’ll see the starter prompts panel with:

• A personalised greeting (“What should we tackle, Aby?”)
• Workspace starter prompts chosen by your workspace admin (e.g. “Next meeting”, “Recent email”, “Today priorities”)
• Familiar starting points derived from your recent conversation history, so you can jump back into a topic with one click

Starter prompts are shortcuts. Clicking one sends that prompt immediately — it’s the same as typing it yourself and hitting Send. They’re a way to discover what PebbleChat is best at in your organisation, and to re-enter conversations you’ve had before.

3. The composer (bottom)

This is where you type. The composer has four rows of UI from top to bottom:

Text input — where you type your message. Supports plain text, rich formatting, and @ mentions. Paste freely — long pastes are accepted as-is.

Left accessories (before the text, reading left to right):

• Gear icon — per-chat settings for chime, background processing, background alerts
• Voice input — press-to-talk or hands-free mode (see Voice)
• Attach files — upload documents, images, spreadsheets (see File Attachments)
• @mentions / Add context — opens the Capabilities & Context menu to reference specific flows, document stores, Past Chats, Agents, Tools, or Rules (see @Mentions & Tools)
• Web search toggle — enable/disable live web search for the next message
• Deep Research toggle — enable a longer-form research mode
• Agentic mode selector — switch between “ask” (standard chat) and other modes
• Thinking effort — for reasoning-capable models, trade latency for reasoning depth

Right-hand controls:

• Model selector — which AI model to use. Defaults to Auto (intelligent routing via PebbleRouter) or your last-used model
• Send button — or press Enter

Below the composer: Starter prompts on the landing view, or the Familiar starting points carousel with recent conversation topics.

4. Top bar (global, not specific to PebbleChat)

Workspace switcher, light/dark toggle, help menu, notifications bell, profile avatar. These are shared across all of PebbleAI — see Collaboration & Profile for the details.

Your first conversation

The best way to learn PebbleChat is to ask it something interesting. Try any of these:

Work-focused prompts

What are the top three risks in rolling out a new CRM to a 500-person sales team, and what should we do about each of them?

Draft a polite but firm reply to a client who has missed three project check-ins in a row. Keep it under 120 words.

Compare the pricing of GPT-4o, Claude Opus, and Gemini Ultra for a use case that sends 10 million tokens per month, and recommend one.

Research-focused prompts (will trigger web search)

What are the latest Anthropic Claude releases in 2026? Summarise the top three with dates.

What's the current state of Australia's electric vehicle market compared to other OECD countries? Include sources.

Task-focused prompts (great for seeing the Activity Stream)

I need to write a project brief for migrating our file shares to SharePoint. Outline the sections, draft the introduction, and list the three most important risks.

I'm presenting to the board on Thursday about AI adoption. Produce a 5-slide outline with talking points for each slide.

What happens next

1. Press Enter or click Send
2. PebbleChat starts working — you’ll see an Activity Stream banner showing the phases (Planning, Web Search, Tool Use, Finalising) when the query needs research or tool calls
3. The response streams in progressively with markdown rendering as it arrives — headings, lists, tables, code blocks all appear before the response is finished
4. When complete, message actions appear: copy, edit your prompt, retry, export to PDF/Word/email

Tips for effective prompts

• Be specific about format. “List 5 bullet points” or “write a two-paragraph email” produces more focused output than “help me write”.
• Include your context. “I’m a marketing analyst at a B2B SaaS company” at the start of a session gives everything that follows a relevant framing.
• Ask follow-ups. Don’t start over — PebbleChat remembers the full conversation. Refine, push back, and explore rather than re-prompting from scratch.
• Experiment with models. The same prompt often produces noticeably different responses on Opus, Sonnet, and Haiku. Try the model selector.
• Save the good prompts. If a prompt works well, ask your workspace admin to add it as a starter prompt so your whole team benefits.

Going deeper — prompt engineering guides

Prompts in PebbleChat work the same way as prompts direct to the model, so the best advice lives at the labs themselves. Each guide has a slightly different voice and covers different patterns — if you’re using a specific model family, start with the matching guide:

• OpenAI — Best practices for prompt engineering — concrete do’s and don’ts, clear examples.
• Anthropic — Claude prompting best practices — XML tags, role prompting, and Claude-specific structure patterns.
• Google — Gemini prompting strategies — prompt design, few-shot examples, and iteration techniques.

On top of what those guides cover, PebbleChat gives you four ways to get leverage without longer prompts:

• Skills — if the same shape of request comes up often, a skill encodes the instructions once and reuses them. Shorter prompts, more consistent output.
• @Mentions — instead of pasting content or explaining where something lives, @mention the flow, document store, or past chat directly.
• Ambient Context — persistent instructions you always want followed (tone, timezone, terminology). Set it once instead of typing it at the top of every session.
• Memory — PebbleChat remembers facts and preferences across conversations. You don’t need to re-explain your role, project, or team every time.

Where to next

• Model Selection — how Auto routing works and when to pick a specific model
• @Mentions & Tools — how to reference your organisation’s flows and document stores
• Advanced Features — background chat, activity stream details, reasoning panels