Skip to main content
The Confidence Agent is an AI assistant built into the Confidence platform. Use it in the Confidence web app to explore and manage your Confidence resources. It can also work with connected data and code sources when your account has access to them.

​
Access the Agent

Open the chat interface in the Confidence web app.

​
Projects

Projects are durable workspaces where you organize related chats and context around a shared goal. When you chat inside a project, the agent automatically has access to the project’s description, connected surfaces, and any attached documents—so you don’t have to repeat background information in every conversation.

​
Create a Project

1

Open the Projects page

Select Projects on the left sidebar.
2

Create a project

Click Create and fill in the project details:
  • Name: A short, recognizable name (for example, “Checkout Redesign Q3” or “Growth Onboarding”).
  • Description: Background context the agent should know when working in this project. Write it as if you were briefing a teammate: what the project is about, what success looks like, and any constraints.
  • Owner: The person or team responsible for the project.
  • Surfaces (optional): The product surfaces this project relates to. The agent uses these to scope queries and pull in relevant experiments and metrics.
You can also attach links during creation.
3

Start chatting

Open the project and use the chat to ask questions or give tasks. Every chat in the project inherits its context.

​
What Goes in a Project

Projects work best when they represent a focused area of work—a feature launch, an investigation, a product area. Within a project you can:
  • Chat: Start conversations that share the project’s context without repeating the same background.
  • Attach documents: Add links, Google Drive documents, or Notion pages to the right-side panel. The agent can read these when they’re relevant.
  • Connect surfaces: Link the product surfaces your project relates to, so the agent can pull in the right experiments, metrics, and flag data.

​
Project Context and One-Off Chats

When you start a chat outside of any project, the agent has access only to the account’s global context and whatever you tell it in that conversation. It doesn’t have the project description, connected surfaces, or attached documents. Use a project when you have an ongoing effort that spans multiple conversations. Use a one-off chat for quick, standalone questions.

​
Global Context

Global context is account-wide background information that the agent always has access to, regardless of which project or chat you’re in. It’s the right place for stable facts about your company that rarely change.

​
Find Global Context

Go to Admin on the left sidebar, then select Context under the Agent section. The page has two parts:
  • Global context: A text field where you write facts, definitions, and conventions you want the agent to know on every task.
  • Context documents: Google Drive documents and Notion pages the agent can read when it needs them.

​
What Belongs in Global Context

Good fit

  • What your product is and who your users are
  • Business model and key metrics definitions
  • Product surfaces and how they map to your apps
  • Internal terminology and naming conventions
  • Experimentation policies (for example, “always include a holdback”)

Not a good fit

  • Task-specific instructions that only apply to one project
  • Temporary information like sprint goals or launch dates
  • Detailed technical specs for a single feature
  • Anything that changes often
Example global context:

​
Add Context Documents

Context documents are Google Drive documents or Notion pages that the agent can read when relevant. Unlike the global context text, the agent doesn’t load these documents into every conversation—it decides when to consult them based on the task.
1

Go to Admin > Context

Open the Context page under Admin > Agent.
2

Add a document

Click Add from Google Drive or Add Notion page. You may need to connect your Google or Notion account the first time.
3

Select the document

Search for and select the document you want to add.

​
How Context Scopes Work Together

The agent combines context from multiple sources on every turn:
  1. Global context: Always available. Account-wide facts and reference documents that apply to every conversation.
  2. Project context: Available when chatting inside a project. Includes the project’s description, connected surfaces, and attached documents.
  3. Conversation context: What you’ve said in the current chat, including any follow-up questions and the agent’s previous responses.
More specific context takes priority when there’s a conflict. If your project description says “the primary metric is checkout completion rate” but you say in the chat “focus on page load time instead,” the agent follows your direct instruction.
Global context is guidance for the agent, not a way to override permissions, safety policies, or access controls. The agent always follows platform rules regardless of what the global context says.

​
Capabilities

The Agent’s available tools can vary by account, permissions, project, and connected services. By default, it can help with these tasks:
  • Feature flags: Find, create, and update flags, variants, targeting rules, clients, and segments.
  • Experiments and rollouts: Create and manage experiments, inspect results, calculate sample sizes, run exploratory analyses, and work with reviews.
  • Confidence resources: Find and summarize metrics, entities, fact tables, surfaces, holdbacks, and other resources you can access.
  • Projects and documents: Use project details and attached context, update projects, and read or edit documents within your current scope.
  • Confidence documentation: Find guidance about Confidence features and SDKs.
  • Data exploration: Generate SQL and charts from your warehouse data and explore metrics conversationally.
Example prompts:
  • “Create a flag for the checkout redesign”
  • “What are the results of the homepage experiment?”
  • “Which surfaces have running experiments?”
  • “How do I set up sequential testing?”

​
Capabilities That Require Setup

Some tools appear only when the required data, connection, permission, or account feature is available:
  • Code exploration requires a connected GitHub account and repository access. The Agent can search and read repositories, inspect pull requests, and find flag usage.
The Agent respects your existing Confidence permissions and connected-service access. It does not grant access to resources you cannot otherwise use.

​
Get the Most Out of the Agent

Be specific. Include flag names, experiment names, or metric names when you can. “What are the results of checkout-redesign?” works better than “show results.”
  • Use projects for ongoing work. Create a project for each feature, investigation, or product area so the agent has the right context from the start.
  • Write good project descriptions. The more the agent knows about what you’re working on, the better it can help. Include goals, key metrics, and relevant flag or experiment names.
  • Keep global context up to date. Review your global context periodically, especially after brand changes, terminology changes, or new product launches.
  • Ask follow-up questions. The agent remembers context within a conversation, so you can refine and build on previous answers.
  • Use it for exploration. Ask “what experiments are running?” or “show the flags” to get oriented before diving in.
  • Try natural language. You don’t need to know exact resource names—describe what you’re looking for and the agent finds it.

MCP Servers

Integrate the Confidence Agent with your IDE

Use AI with Confidence

Set up MCP for AI-assisted development

Use Review Agents

Set up AI agents that review experiment designs