Confluence

Connect an Atlassian Confluence Cloud workspace so Cardinal’s AI agent can search and read your team’s pages on demand — runbooks, design docs, onboarding guides, knowledge-base articles, and more.

Overview

With the Confluence integration, the AI agent can:

• Find the right page — search across your spaces using Confluence’s native query language
• Read page content — fetch a page’s body in clean Markdown for the agent to summarize, quote, or extract from
• Track changes — list a page’s version history when the user asks who edited what and when
• Discover what’s available — enumerate spaces the bot has access to

Pages are fetched fresh on every turn, so edits in Confluence are immediately reflected in the agent’s answers — there is no sync to wait on, no index to rebuild.

Capabilities

Configuration

Settings

Credentials

Authentication uses HTTP Basic Auth.

Prerequisites

• An Atlassian Confluence Cloud workspace. (Server / Data Center is not currently supported.)
• A service account with read access to the spaces you want the agent to use. The agent can only see what the service account can see — restricted spaces invisible to the service account are also invisible to the agent.

Setup

1. Create an Atlassian API token at id.atlassian.com → Security → API tokens . Use a descriptive label such as cardinal-bot.
2. In Cardinal, navigate to Settings → Integrations → New Integration and select Confluence.
3. Fill the required fields:
   • Base URL — your workspace root.
   • Service Account Email — the email tied to the API token.
   • API Token — paste the token value.
   • Spaces include / Spaces exclude — optional, see below.
4. Acknowledge the data-handling notice shown on first save.
5. Click Test Connection to verify Cardinal can reach your workspace and the credentials are valid.
6. Click Save.

Working with multiple Confluence workspaces

If your organization runs more than one Confluence Cloud workspace — for example after an acquisition or to separate product lines — you can configure each as a separate Confluence integration. Each integration has its own slug, base URL, and service account credentials. The agent picks the right workspace based on conversation context, and you can list configured workspaces from inside any chat.

Choosing which spaces the agent uses

For a focused, predictable agent experience, scope the integration to the spaces you actually want it to draw from.

• Spaces include narrows the agent to a specific list of space keys. If left empty, the agent can search every space the service account can read.
• Spaces exclude removes spaces from the agent’s reach. Exclude takes precedence over include, so a key on both lists is excluded.

A good default is to exclude any space that contains credentials, HR/legal/finance content, security policies, onboarding material, or anything else you wouldn’t want quoted back in a chat answer. Many organizations start with an exclude list like:

SECRETS,VAULT,CRED,HR,LEGAL,FIN,FINANCE,PEOPLE,ONBOARDING,SECURITY

Adjust to match your space-naming conventions. The integration’s Included spaces preview panel in Cardinal shows which spaces the agent can actually see after your include/exclude lists are applied — a quick way to confirm you’ve scoped it the way you intended.

Privacy and data residency

Cardinal runs language-model calls inside your own AWS account using Amazon Bedrock, so the contents of your Confluence pages stay within your AWS environment on the LLM path — they are never sent to a third-party model provider.

Any user with agent access in your Cardinal organization can ask the agent about any page the service account can read. Use Spaces exclude to keep sensitive spaces out of the agent’s reach, and grant the service account only the Confluence permissions it actually needs.

What this enables

Once configured, you can ask the agent questions like:

• “What does the SSO bootstrap runbook say about rotating the IdP signing key?”
• “Summarize the design doc for the payments reconciler.”
• “Find the on-call escalation policy for the data platform team.”
• “Who edited the incident-response page most recently?”