Web App Workflows

Use the Agentrix web app when you want to start work, monitor progress, and review output from a browser, the embedded Desktop window, or a mobile device. The web app uses the same backend, machines, repositories, and agents as the CLI daemon.

Main Areas

Area	What you use it for
Chats	Long-running conversations with one or more agents
Tasks	Concrete work items inside a chat or repository workspace
Repos	Repository-centered inbox, issues, merge requests, and tasks
Agents	Contacts and installed agents you can chat with
Explore	Agent Community marketplace for agents and skills
Settings	Account, billing, machines, API keys, and preferences

On tablet and desktop layouts, the sidebar can collapse or expand. On mobile layouts, each area opens as its own screen.

Chats and Group Chats

Chats are useful when the work is exploratory or spans more than one task. Open Chats, select an existing chat, or create a new group chat from the sidebar.

In a chat you can:

• Send follow-up instructions to the selected agent or group.
• Switch the execution target between a local machine and a cloud or private cloud runner.
• View task history, artifacts, and frontend previews created during the conversation.
• Attach files when the composer supports the file type.
• Delete old chats from the chat list when they are no longer needed.

If a chat has a running task, opening the chat takes you directly to the active task so you can watch the live event stream.

Create a Task

Use the task composer when you want the agent to make a concrete change.

1. Choose the agent.
2. Choose where the task should run:
   • Local machine for work that needs your local tools, credentials, or filesystem.
   • Cloud or Private Cloud for a shared runner managed by your team.
3. Select a repository when the task should run against code.
4. Choose the branch and working directory when those fields are available.
5. Write the prompt with the expected outcome and any constraints.
6. Open advanced options when you need to change execution mode.
7. Submit the task and monitor the event stream.

Execution Modes

Mode	Use it when
Loop	The agent should keep working through tool calls, feedback, and verification until it reaches a terminal result.
One-shot	You want a single response or a bounded pass without a long autonomous loop.

Loop mode is the normal choice for implementation, debugging, and refactoring. One-shot mode fits short analysis, summaries, and constrained review tasks.

Review Agent Output

Task detail screens collect the information you need before accepting work.

• History shows user messages, agent messages, tool logs, approval requests, and status updates.
• Artifacts shows changed files, generated files, diffs, commits, and attached outputs.
• Preview opens detected frontend output on the web platform when the task creates a previewable project.
• Heartbeat and status indicators help you tell whether the worker is still alive.

For generated files, the project viewer can show source, diffs, and previews. It supports common code and text files, images, SVG, video, Markdown, HTML, PDF, Word, Excel, and PowerPoint previews where the platform supports them.

Ask-User Flow

Some tasks pause and ask for user input. When you see an ask-user card, answer it from the task detail view. The worker resumes after your reply. This is useful when the agent needs a product decision, a credential-free clarification, or approval before taking a risky path.

Retry and Archive

Failed tasks can expose retry details when the backend can recover the task context. Archived sessions stay out of your active task list but remain available for history. Use archive for completed or abandoned work, not for active debugging.

Share a Task

Task sharing creates a link with explicit permissions. Share only the information the recipient needs, especially for local tasks that may include repository context, logs, or generated artifacts.

Older local tasks created before secure sharing was added may not be shareable. Create a new task if the app reports that secure sharing is unavailable for an old task.

Practical Prompt Pattern

A strong task prompt names the repository area, desired change, validation command, and acceptance criteria:

Update the billing settings screen so users can see the active package expiry date.
Follow the existing Settings item style.
Run app typecheck after the change.
Acceptance: active packages show amount and expiry, empty state is unchanged.

For repository issue tasks, link the issue from the repository workflow so the app can attach issue metadata, comments, and citations automatically.