// Guide

Use ArtifactBridge AI

ArtifactBridge AI is the agent available in the browser workspace. Use it to work with accessible documents and discussions without starting a local coding tool.

Start a conversation

1. Open the Agent launcher or an Ask agent action.
2. State the result you need and name the relevant material.
3. Use the available mentions to identify specific workspace context.
4. Send the request. Inspect the progress, result, and linked evidence.

From a Room, Send to agent → Send to ArtifactBridge AI starts work with that Room’s context. Existing conversations can be opened from the agent conversation history. A new conversation starts its own thread of work.

Ask a specific question

Specify the source, output, and decision.

For local files that have not been imported, use your local agent. The built-in browser agent does not gain access to your computer merely because you mention a file path.

Review agent actions

The agent can read available context, prepare documents and proposals, and use enabled connectors. Follow its progress and open the source links when a result needs verification. When it asks for an action approval, inspect what will run and where.

External actions such as commenting on a GitHub issue, commenting on a Linear issue, or creating a Linear issue use the configured connection and require the approval presented by the product. Accepting a document proposal, publishing content, and changing access are also explicit decisions. A read permission is not permission to perform all actions.

Configure the workspace agent

Company owners manage Settings → Agent.

• Connectors: enable supported GitHub and Linear reads and configure their scope.
• Slack: connect the company bot and configure its available sources and behavior.
• Guidance: state workspace instructions the agent should follow.
• Persona: set the agent’s displayed name and voice.

Guidance and persona shape behavior and language. They do not grant permissions or remove review requirements. Put reusable task instructions in a Skill and repeated tasks in a workflow.

Inspect agent activity

Open Settings → Logs → Agent activity to inspect the tool calls you are permitted to see. Use the Inbox for approvals and failures that need you. If a result is incomplete, ask a follow-up that names what is missing and which sources to check.

Collect and review knowledge

Open Settings → Agent → Harvest for knowledge collection. A harvest scans configured sources for useful findings. The run list shows success, skipped work, or failure. The findings list shows open changes, accepted findings, and dismissed findings. Open a finding in the Inbox to review it; a proposed finding is not an approved fact.

Workspace members can read the category list. An owner or administrator can add, edit, reorder, archive, or restore categories. Category order sets priority: the classifier reads from top to bottom. Archiving keeps the category record; it does not delete existing documents. Use the category field on a supported document to classify it. Categories do not change access. Detailed harvest activity is available to owners when the workspace has the required bot connection.

Access and limits

The agent works within the current workspace and the connected member’s allowed access. Connector scope, missing permissions, provider failures, and plan limits can restrict a task. Read the stated reason and correct it before retrying. The app explains when a feature is unavailable for the workspace’s current plan.