Quick Start

Quick Start

Welcome to PebbleAI — your AI-powered workspace where conversation, automation, and observability come together. This guide gets you from logged-in to productive in about five minutes, starting with PebbleChat, the powerhouse of the platform.

Prerequisites

  • A PebbleAI account on your organisation’s PebbleAI install (e.g. demo.pebblecloud.io)
  • A working web browser
  • Your organisation admin has provisioned at least one model for you (most installs do this automatically — if not, see User Settings → Models)

You do not need an API key from any LLM provider. Your organisation has already set up shared credentials, so the moment you log in you can start chatting.

1. Open PebbleChat

After logging in, click PebbleChat in the left sidebar — it’s the first item, with a chat icon. You land directly in the chat interface, ready to type. There is no “open chat” or sub-menu step.

PebbleChat landing page with starter prompts

You’ll see:

  • A personalised greeting (“What should we tackle, Aby?”)
  • Workspace starter prompts — chips your workspace admin has curated for common tasks
  • The composer with an input box, attachment button, voice button, and a model selector

2. Send your first message

Type a question and press Enter, or click one of the starter prompts. Some prompts that show off PebbleChat well:

Summarise the three biggest risks in rolling out a new CRM to a 500-person sales team, and recommend a mitigation for each.
What are the latest Anthropic Claude releases in 2026? Include dates and a one-line summary of each.
I'm presenting to the board on Thursday about AI adoption. Produce a 5-slide outline with talking points for each slide.

The response streams in progressively as formatted markdown — you can start reading before it finishes. For research-heavy questions you’ll see a Research & Tools activity stream banner above the response showing the model planning, searching, and synthesising in real time.

Conversation with Research & Tools activity

When the response is complete, hover over it to see message actions: copy, edit your prompt (and resend), retry, and export to PDF / Word / email / markdown / HTML.

No thumbs up / down here. PebbleChat doesn’t have per-message rating buttons. To send feedback, use the Help menu → Submit Feedback in the top bar — it’s a draggable modal that supports up to five screenshots and lands directly in the engineering team’s Jira queue. See Help Menu & Feedback.

3. Choose a model — or let PebbleChat choose

The model selector sits to the left of the Send button. By default it shows Auto — that means PebbleChat is using your organisation’s PebbleRouter to intelligently pick the best model for each message based on cost, capability, and availability.

Model selector showing Auto and Bedrock models

In most installs, the available models come from AWS Bedrock (Claude Opus, Sonnet, Haiku) — Bedrock keeps your data inside your AWS region (e.g. ap-southeast-2 for Australia), which is important for compliance, data residency, and contractual coverage. Your organisation may also expose direct OpenAI, Anthropic, or Google models depending on its setup.

When to switch from Auto to a specific model:

  • You started with a fast model and want a more capable one for the next, harder question
  • You want deterministic behaviour — the same model every time, regardless of routing
  • You want to compare how Opus vs Sonnet vs Haiku handles the same prompt

Switching mid-conversation is allowed — the full conversation history goes with the new model.

See Model Selection for the full picture.

4. Reference your organisation’s tools and knowledge

PebbleChat isn’t just a chat with a model — it’s a chat with everything your organisation has built. Click the @ button in the composer (or type @) to open the Capabilities & Context menu:

Capabilities & Context menu

You can reference:

  • Document Stores — knowledge bases your team has indexed for RAG
  • Past Chats — earlier conversations, used as context for the current one
  • Assistants — pre-built single-agent helpers
  • Agents — multi-step Agentflows from PebbleFlows
  • Tools — web search, MCP integrations (Microsoft 365, Jira), custom HTTP tools
  • Rules — admin-curated prompt rules

Or just leave it on Auto and let PebbleChat discover the right asset automatically — it uses embedding similarity to find the most relevant flows, agents, and document stores in your workspace and pulls them in without you naming them. You’ll see what was discovered in the Research & Tools activity stream.

See @Mentions & Tools for the full feature.

5. Use power features as you need them

PebbleChat has several features designed for serious work. You don’t need them right away, but they’re there when you do:

  • Background Chat — long-running conversations keep processing on the server when you navigate away. A spinner in the sidebar shows what’s still running, and a notification fires when it’s done. Toggle it on via the gear icon in the composer.

  • Activity Stream — the Research & Tools panel above each response shows exactly what PebbleChat did to produce it. Click any step to verify sources, see tool inputs/outputs, and trust the response.

  • Thinking & Reasoning — for reasoning-capable models (Claude Opus, o1, o3), an expandable Thinking panel shows the model’s chain-of-thought before the final answer.

  • Voice Input — talk to PebbleChat in press-to-talk or hands-free mode. Full feature parity with typing.

  • File Attachments — upload PDFs, Word documents, spreadsheets, code, emails, images. PebbleChat reads and reasons about them in the conversation.

  • Export — every response can be exported as PDF, Word, email, markdown or HTML. The “Submit Feedback” modal in the help menu is your channel for telling the team what’s working.

6. Building Custom AI Workflows with PebbleFlows

Once you’re comfortable in PebbleChat, the natural next step is to build reusable assets that PebbleChat can call on — flows, agents, document stores, custom tools. That’s where PebbleFlows comes in.

PebbleFlows isn’t a separate product from PebbleChat; it’s the thing that makes PebbleChat more powerful. Every flow you build in PebbleFlows becomes available to PebbleChat in three ways:

  1. As an @-mention — users type @my-flow to invoke it
  2. Via asset discovery — if marked discoverable with a clear description, PebbleChat finds it automatically
  3. Via API — call it from your own backend, mobile app, or CI pipeline

When to build something in PebbleFlows

  • Repetitive tasks that need automation
  • Multi-step processes where you want the same flow every time
  • Custom RAG pipelines over your team’s documents
  • Integrations with services PebbleChat doesn’t have built in
  • Workflows you want everyone in the org to be able to invoke from PebbleChat

Start with Agentflows, not Chatflows

PebbleFlows ships with three builders:

  • Agentflows (recommended) — multi-step, multi-agent, branching, looping. The most powerful builder and where new work should start. Supports everything the others do, plus more.
  • Chatflows — linear single-agent chat flows. Still supported for existing flows, but new work should use Agentflows — Chatflows are being gradually phased out in favour of Agentflow V2.
  • Assistants — beginner-friendly single-purpose helpers. Good for prototypes.

To start building, click PebbleFlows in the sidebar, then Agentflows → Add New. Drag nodes onto the canvas, connect them, configure each one, and save.

The full PebbleFlows documentation lives at PebbleFlows — start with the Using Flowise → Agentflow V2 section.

7. Understanding Your AI Usage with PebbleObserve

Once people in your organisation are using PebbleChat and PebbleFlows in earnest, the obvious next question is: what’s it costing? what’s it doing? is it working? That’s PebbleObserve.

Every PebbleChat conversation, every PebbleFlows agent run, and every API call is automatically traced — no instrumentation, no setup, nothing for users to enable. You can see:

  • Cost — dollar-accurate spend per user, per workspace, per model, per tool
  • Activity — what’s being asked, by whom, with what tools
  • Performance — latency, error rates, retries
  • Quality — set up evaluations to measure response quality systematically

Personal usage lives at My Usage (in your profile fly-out). Organisation-wide usage lives at Admin → Usage and Admin → Logs — only org admins can see those views.

You’re set

That’s the five-minute tour. Where to go next depends on your role:

If you’re a regular user

If you’re going to build flows

If you’re an admin

Tip: bookmark PebbleChat — it’s where you’ll spend most of your time, and everything else exists to make it more powerful.

WelcomeWhat is PebbleAI?