Skip to content

Conversations

Conversations are the core interaction model in ReArch. Each conversation connects a user to an AI coding agent running inside a dedicated, sandboxed Docker container linked to a specific repository. Through conversations, you can describe tasks in natural language, watch the agent work in real time, review changes, and ship pull requests.

There are two ways to start a new conversation:

Navigate to Conversations > New (or press the “New conversation” button in the sidebar). You will see a grid of your enabled repositories. Select the repository you want to work with, then click Start Conversation.

A new conversation is created and a background job begins provisioning a sandboxed environment. The container clones your repository, boots the AI agent, and becomes ready for interaction. You can monitor the container status in the session sidebar.

You can create a conversation directly for a specific repository by navigating to /start#<repository-name>. ReArch will look up the repository by name, verify it is enabled, and automatically create the conversation. This is useful for integrations, bookmarks, or shared links within your team.

Each conversation follows a predictable lifecycle:

  1. Created — The conversation document is created and a container setup job is enqueued.
  2. Starting — A Docker container is provisioned with your repository and AI agent.
  3. Running — The environment is ready. You can send messages, and the agent can read, modify, and run code.
  4. Stopped — The conversation has been deleted or the container has been cleaned up.

Container provisioning is automatic. You do not need to manage Docker containers directly. ReArch handles creation, health checks, and teardown.

Once the environment is running, type your message in the input field at the bottom of the chat interface. You can:

  • Describe tasks, bugs, or features in natural language.
  • Attach files for additional context.
  • Select a specific AI model (e.g., anthropic/claude-sonnet-4-20250514) from the model selector.
  • Select a specific agent persona (e.g., code, build) from the agent selector.

Messages are streamed in real time via Server-Sent Events (SSE). As the agent works, you will see its reasoning, tool calls (file edits, command execution), and responses appear progressively.

During a task, the agent may ask clarifying questions — for example, to confirm a destructive action or choose between implementation approaches. These appear inline in the chat. You can reply with an answer or dismiss the question.

You can edit a previously sent message to re-run a prompt with different instructions. The agent will re-process the updated message from that point.

The session sidebar (right panel) provides real-time information about the active conversation:

SectionDescription
Container StatusCurrent state of the environment (starting, running, stopped, error).
RepositoryThe linked repository name.
Context UsageA progress bar showing how much of the AI’s context window has been consumed.
Session CostAccumulated AI usage cost for this conversation in USD.
ParticipantsAvatars of all users who have sent messages in this conversation.
ServicesQuick-access buttons to open VS Code (code-server), the running application preview, or other services exposed by the container.
Changed FilesA live list of files modified by the agent, with added/removed line counts.

After the agent has made changes, you can review and ship them directly from the conversation.

The session sidebar displays all modified files. You can view a full git diff of the working tree to inspect exactly what the agent changed before committing.

Click Conclude change in the session sidebar to open the commit and push workflow:

  1. Stage changes — Review the files to be committed.
  2. Write a commit message — Provide a descriptive commit message.
  3. Push — ReArch creates a branch, stages, commits, and pushes the changes inside the container.

After pushing, you can create a pull request directly from ReArch. The PR creation form lets you:

  • Set a title and description.
  • Select reviewers from your Bitbucket workspace members.
  • Choose the source and target branches.

Created pull requests are tracked on the conversation and visible in the session sidebar.

The sidebar displays all your conversations sorted by most recently updated. Each entry shows:

  • Title — The conversation name (defaults to “New Conversation”).
  • Time — How long ago the last activity occurred.
  • Repository — Which repository is linked.
  • Status indicators — A spinner when the agent is busy, and an unread badge when new activity has occurred.

Use the search field at the top of the sidebar to filter conversations by title. The search is case-insensitive and matches partial text.

Right-click a conversation in the sidebar (or use the context menu) and select Rename to change its title.

Right-click a conversation and select Delete. This performs a soft delete — the conversation is hidden from the sidebar and a background job cleans up the associated Docker container. Soft-deleted conversations retain their data and are excluded from normal queries.

Deletion is soft by default. The conversation data is preserved after deletion. The associated container and its runtime state are permanently removed.

Conversations update in real time via WebSocket. You do not need to refresh the page to see changes. The following events are pushed automatically:

EventDescription
Busy / IdleIndicates whether the AI agent is currently processing a prompt.
Session InfoUpdated context usage and cost after each interaction.
Job StatusProgress of container setup or teardown jobs (active, completed, failed).

When a user sends a message in a conversation, they are automatically added to the participants list. Participants are displayed as avatars in the session sidebar. This provides visibility into who has contributed to a conversation, which is useful for team collaboration.

Use the Command Palette to quickly navigate between conversations:

  • Open the palette with the keyboard shortcut.
  • Select Switch conversation to jump to any existing conversation.
  • Select New conversation to start a fresh one.
  • Select Switch model to change the AI model for the current session.