Skip to main content

Chat

Chat is the primary way to interact with Revelry Ai. Powered by MCP (Model Context Protocol), you can have multi-turn interactions that dynamically gather information and help you create artifacts.

How chat works​

The chat interface enables iterative, context-rich interactions with AI:

  1. Start a chat - Ask questions, request analysis, or describe what you want to create
  2. AI uses tools dynamically - The AI calls tools as needed to gather information (documents, GitHub data, etc.)
  3. Refine iteratively - Continue the chat to clarify, expand, or revise
  4. Artifacts are created - The AI creates artifacts directly using MCP tools, or you can manually export specific messages

This is a longer loop than a single prompt and response. The AI can recursively gather information until it has what it needs to give you the best answer.

Key features​

Flexible starting points​

You can start a chat in several ways:

  • Type a question or describe what you want
  • Run a Skill from your Skill Library
  • Open a chat with an Agent that's set up for a specific job
  • Attach specific files, or turn on Knowledge so the AI can search your Knowledge Base
  • Use Web Search, which is on by default, for current information

Multi-turn interactions​

  • Build context - Refine over multiple messages
  • Streaming responses - See AI responses in real-time as they're generated
  • Persistent sessions - Return to previous chats anytime

Sending messages​

  • Desktop - Press Enter to send. Shift+Enter adds a line break
  • Phones and tablets - Pressing Enter adds a line break. Tap the send button to submit

MCP tool integration​

The AI can dynamically call tools during chat to:

  • Query documents - Search your Knowledge Base for relevant information (when Knowledge is on)
  • Access external sources - Pull data from GitHub, filesystems, and other connected tools
  • Search the web - Find current information online (when Web Search is on)
  • Examine and refine - Look at results and decide what additional information is needed

You control which tools are active and what context is available for each chat.

Source attribution​

When the AI uses tools, you'll see:

  • Which tools were called
  • What information was retrieved
  • Where the information came from (citations)

This transparency helps you understand and trust the AI's responses.

Chat Recommendations​

When you start a new chat, the landing page can greet you with Chat Recommendations: personalized suggestions for your next conversation, generated from your recent activity (your chats, artifacts, documents, and the Skills and Agents available to you). Each recommendation is a ready-to-send prompt, sometimes pointing at a specific Agent or Skill. Clicking one prefills the composer and enables the tools the task needs, and Show new ideas regenerates the set.

Recommendations are cached, so they appear instantly and only change when your activity does or you ask for new ones. Everything they draw on is permission-filtered, so they never surface activity you can't already see.

You can turn recommendations off under Chat Preferences on your Profile page, and turn them back on the same way.

Note: Chat Recommendations are behind a feature flag. If you don't see the Chat Preferences setting, contact your Revelry Ai team to enable.

Starting a chat​

  1. Navigate to Chat in the left navigation
  2. Select your Project (determines context and available data)
  3. Choose your starting point:
    • Start typing directly for a freeform chat
    • Type / to pick a Skill from your Skill Library and begin with a template
    • Attach files or documents from your Knowledge Base for context

The chat box stays compact and grows as you type.

Using Skills​

You can start a chat with a Skill from your library by typing / in the chat box. The typeahead searches both Organization Skills and Revelry Ai Skills. Pick one and it starts right away, and if the Skill needs information from you, the AI asks for it immediately.

This gives you a structured starting point while maintaining the flexibility of chat. You can also open a Skill from the Skill Library and click Use Skill to run it in chat.

Skill slash commands​

You can also invoke a Skill directly from the composer by typing /. A typeahead suggests Skills by their slug as you type. Navigate it with the arrow keys and press Enter to pick one:

  • A Skill you pick starts immediately, without the AI searching for it first. A Skill with no required inputs (or a single input) runs right away, and the rest of your message is used as the input.
  • A Skill with multiple inputs is handed to the assistant, which asks for what it needs straight away and runs the Skill.
  • Each Skill's slug is auto-generated and can be edited by your organization. If what you type doesn't match a slug, the message is sent to the AI as-is.

See Executing Skills → Slash commands for the full behavior.

In-context or separate Skill runs​

A Skill invoked during a conversation can run in one of two ways, and the assistant picks based on the task (or you can ask for one explicitly):

  • In context: the Skill's instructions load into your current conversation, so it sees everything you've discussed and the work happens inline, driven by the conversation.
  • Separately: the Skill runs in an isolated session with its own model and tools and returns only its final result. Better for large outputs or tasks that don't need your conversation's context.

See Executing Skills for details on both modes.

Attaching context​

Before or during your chat, you can attach specific context:

Attach Files:

  • Upload files directly to the chat
  • Select documents from your Knowledge Base
  • Attach specific files for focused context

The Context button:

  • When you add more than one attachment, tool, integration, or filter, they're grouped under a single Context button below the chat box
  • Click it to review everything attached to the conversation and remove one item or all of them

Knowledge Search:

  • Turn on Knowledge in the Chat Tools menu to let the AI search your whole Knowledge Base
  • Attach specific files instead, or as well, when you want focused context

Note: Knowledge is off by default. Some organizations have it on by default for every new chat. Contact your Revelry Ai team to enable.

Web Search:

  • Web Search is on by default, for current information or topics outside your Knowledge Base
  • Turn it off in the Chat Tools menu to keep answers to your files and Knowledge Base

Chat Tools menu​

Use the Chat Tools menu (the + button next to the message input) to choose which tools are active. That covers the built-in tools (Knowledge, Artifacts, Web Search, Workspace Agents) and external integrations (GitHub, Pipedrive, HubSpot, and others). The model you choose and the tools you leave on are remembered as defaults for future chats until you change them. See MCP Tools: Chat Tools menu.

Tip: Start with specific attachments for focused chats, or turn on Knowledge for broader exploration.

When to use Chat, Skills, Agents, or the API​

Use free-form Chat when you:

  • Want to explore ideas through open-ended AI interaction
  • Need the AI to dynamically gather information
  • Want to refine outputs iteratively
  • Need to provide specific context files or enable flexible search
  • Prefer flexible interaction over a fixed setup

Run a Skill when you:

  • Have a well-defined, repeatable task
  • Want structured, packaged work, either inline in your conversation or as a separate self-contained run
  • Note: Skills run inside a chat (from the in-chat picker, a slash command, or Use Skill in the Skill Library) or programmatically through the API

Chat with an Agent when you:

  • Want a conversation that's already configured for a specific role or job
  • Want the AI to follow a consistent persona and instructions across the whole session
  • Want a curated set of Skills available without picking them each time
  • See Agents

Use the API or Workflows when you:

  • Need programmatic artifact generation (custom apps)
  • Want automated or scheduled generation
  • Need a fixed, multi-step pipeline that runs without a person in the loop

Creating artifacts from Chat​

AI-created artifacts (primary)​

In most conversations, the AI creates artifacts for you using its artifact MCP tools as part of the conversation. It writes them directly when the content is ready, or hands off to a background subagent when it needs to research first. You describe what you need, and the AI gathers information and generates the artifact.

Manual export​

You can also manually save a specific chat message as an artifact:

  1. Click Save as artifact on the message you want to keep
  2. Optionally choose an Artifact Type and a Category
  3. The message content becomes a saved artifact

This is useful when you want to save a specific message exactly as-is.

Learn more: Exporting from Chat

Archive chats​

You can archive chats you no longer need:

  1. Find the chat on the Chat History page
  2. Click the archive option on the chat
  3. The chat is removed from the list

Archived chats are not deleted, all data is preserved. See Chat History for details.

Chat compaction​

During long conversations, Revelry Ai automatically summarizes older messages to keep the conversation manageable. You'll see compaction status cards inline in the conversation when this happens.

Learn more: Chat Compaction

Thinking blocks​

When a model has reasoning turned on (Adaptive Thinking on newer Claude models, or a reasoning setting on other providers), you may see thinking blocks in the AI's response. These show the model's reasoning process before it produces its final answer.

Tool execution cards​

When the AI calls MCP tools during a conversation, you'll see individual tool execution cards for each tool call. Click a card to see the full details of what was sent to and returned from the tool.

Skill runs are named by their slash command: a card reads "Completed /create-artifact" rather than a generic label, and its expanded view links to the Skill's detail page. Failed executions expand to show the actual error message, and canceled executions are labeled as canceled so they don't read as failures.

Chat History​

All your chat sessions are saved and accessible from Chat History. You can:

  • Return to previous chats
  • Continue where you left off
  • Review past interactions and exported artifacts
  • Archive conversations

Next steps​