PebbleChat
PebbleChat is the powerhouse of PebbleAI. Not a chatbot that answers questions and stops — an AI agent you work with to get real work done. It talks to your workflows, your document stores, your tools, and your data. It researches, plans, acts, and explains what it did.
Everything else in PebbleAI exists to make PebbleChat more capable. PebbleFlows builds the reusable building blocks it calls on. PebbleObserve measures how well it’s performing. PebbleAgents (coming soon) will let it run autonomously in the background. But the conversation in PebbleChat is the place work actually happens.
What makes PebbleChat different
Your conversations stay inside your organisation. Unlike public AI chat products where every prompt leaves your walls, PebbleChat runs inside the boundary your organisation controls. Model calls route through your own credentials — typically AWS Bedrock in your chosen region (e.g. ap-southeast-2 for Australia) — so data residency, contractual coverage, and compliance regimes like IRAP PROTECTED are respected automatically. Take control of your AI usage and safely leverage the competitive advantage AI brings, without handing your organisation’s data to a third party.
One place for all your AI. PebbleAI’s core promise is consolidation: instead of your team spreading sensitive work across a dozen external AI tools, PebbleChat gives you one sanctioned place to do it. Admins see who is using what, finance sees the cost, and security sees the boundary. Everyone gets the productivity; no one loses the oversight.
It’s connected to your organisation’s knowledge and tools. Type @ to reference a specific document store, agent, flow, or skill. Or just ask — PebbleChat automatically discovers the relevant assets your colleagues have built and uses them on your behalf. No plumbing, no lookup tables, no “which bot do I ask?”.
It runs code, not just talks about it. Skills (Claude Code-compatible prompt templates) can include Python that executes in a platform-hosted, gVisor-isolated code execution sandbox. Ask for a branded PDF, a chart, a spreadsheet, a data analysis, and you get the actual file — uploaded to your user file store and linked in the chat. No desktop handoff, no manual copy-paste. Claude, OpenAI, and Copilot don’t do this inside enterprise chat.
It thinks before it answers. For complex queries, PebbleChat plans its approach, runs web searches or tool calls, synthesises the results, and only then writes the final response. You can watch it work in the Activity Stream, expand reasoning panels for reasoning-capable models, and see exactly what sources it used.
It remembers what you tell it. Personal memory, personal ambient context, workspace context, and organisation-level instructions all stack together — so PebbleChat arrives at your first message of the day already knowing your role, your preferences, and what you’re working on. All of it stored against your organisation, not a public account.
It exports into your workflow. Responses download as PDF, Word, email, and other formats. Nothing is trapped in the chat.
It works in the background. Start a research task, navigate away, come back ten minutes later. PebbleChat keeps going and fires a notification when it’s done.
It listens as well as it reads. Real-time voice input with press-to-talk or hands-free mode. Everything you can do by typing, you can do by speaking — including the tool calls. Speech-to-text and text-to-speech run through providers your admin chooses, so voice stays inside the same boundary as text.
Every interaction is observable. PebbleChat is instrumented end-to-end by PebbleObserve. Admins can see cost, tool usage, latency, and quality metrics without anyone having to enable tracing. You get the productivity benefits of AI without losing visibility into what’s happening.
The PebbleChat surface
Here’s what’s on screen when you open PebbleChat:
| Area | What it is |
|---|---|
| Left sidebar | Conversation history (toggle open/closed). Search, new chat, date groupings. |
| Top bar | Workspace switcher, light/dark toggle, help menu, notifications, profile avatar — covered in Collaboration & Profile. |
| Chat area | The conversation itself — your messages, the AI responses, the activity stream. |
| Composer | The input area where you type, attach files, use voice, pick the model, and adjust per-chat settings via the gear icon (chime, background processing, background alerts). |
In this section
| Page | What’s inside |
|---|---|
| Getting Started | UI walkthrough, your first conversation, starter prompts |
| Model Selection | Auto routing, Bedrock and other providers, switching models mid-chat |
| Conversations & History | Sidebar, search, renaming, background chat indicators |
| Messages & Responses | Streaming, thinking panels, message actions, exporting |
| @Mentions & Tools | Referencing flows, Past Chats, MCP tools, asset discovery |
| Skills | Reusable prompt templates with optional Python code that run in the sandbox |
| File Attachments | Supported formats, how attachments are used in context |
| Voice Input | Press-to-talk, hands-free, STT/TTS |
| Advanced Features | Background chat, activity stream, context window, reasoning panels |
| Chat Settings | The composer gear icon — chime and background toggles |
Quick start
- Click PebbleChat in the sidebar (it’s the first item)
- Start typing in the composer, or click one of the starter prompts that appear on the landing page
- Press Enter or click Send
- Watch PebbleChat plan, research, and respond
If you’re brand new, Getting Started walks through the interface element by element.