Chat Interface

The chat interface is your primary way to communicate with Codex in a workspace thread. The thread runs on the selected registered host and uses the workspace folder, Git branch, terminal, and files.

​

Message Types

The chat displays different types of messages, each with distinct styling:

​

User Messages

Your messages appear with an edit option. Click the pencil icon to modify and resend a message.

​

Codex Messages

Codex responses are rendered with full markdown support, including:

• Code blocks with syntax highlighting
• Tables and lists
• Links and formatted text

​

System Messages

Informational messages from the system, such as model initialisation or status updates.

​

Error Messages

Error messages appear in red with a warning icon. Click to expand and see full error details.

​

Chat Input

​

Writing Messages

The chat input supports rich text editing:

• Bold - Cmd/Ctrl + B or click B
• Italic - Cmd/Ctrl + I or click I
• Underline - Cmd/Ctrl + U or click U
• Strikethrough - Click S
• Code - Click the code button or wrap with backticks

​

Attaching Images

You can attach images to your messages:

1. Click the attachment button (paperclip icon) in the toolbar
2. Drag and drop images directly into the chat
3. Images are uploaded and referenced in your message

​

File Mentions

Type @ to mention files from your repository. The chat provides typeahead suggestions showing:

• Filename in bold
• Full path below each file
• Matches update as you type

Select a file to include it as context for Codex.

​

Sending Messages

​

Send Actions

​

Execution States

The chat interface shows different states:

​

Scenario Workflows

Installed scenarios can appear in the chat toolbar as workflow actions. They start a saved linear chain of agent prompts, human gates, and scenario calls using the installation’s runtime mappings and default launch inputs. Before a scenario appears here, an admin must install it in Settings → Scenarios, configure runtime mappings, enable it, and turn on Toolbar. When you start one from chat, Jaidu shows a launch preview with steps, inputs, runtime requirements, and backend preflight issues.

​

Model and Reasoning

​

Choosing a Model for the Next Turn

Use the compact model selector to choose the model and reasoning level for a Codex turn. The selector is available after you select a registered host while creating a workspace, and in existing or new threads inside a task workspace. It is not shown in Main Chat. In a workspace thread, it appears before Runtime resources. The selected registered host supplies the available models, each model’s reasoning options, and the effective Codex model and reasoning defaults for the selected profile and project. The focused workspace controls do not include a Default choice: they show concrete values for the next turn. When you choose a model, Jaidu also selects a compatible remembered or default reasoning level. You can then change the reasoning level from the same control. Your selection applies to:

• a direct send;
• a queued follow-up;
• a message retry or edit.

During workspace creation, Jaidu retains the selection in the creation draft and sends it with the initial create-and-start request. In a workspace thread, the model and reasoning remain selected after you submit a message. Changing them while Codex is running does not switch the active turn; it changes only a later send, queue, retry, or edit. A queued follow-up keeps the model and reasoning captured when you queued it, even if you change the composer before that follow-up starts. Jaidu does not persist the provisional result while the host catalogue is loading. When final discovery arrives, Jaidu fills a missing selection with the host’s effective defaults. Older hosts that do not report defaults use a valid advertised model and compatible reasoning fallback instead. If the host cannot read its Codex configuration, it finalises that concrete fallback without clearing your message or draft. A saved non-null model or reasoning selection takes precedence over host defaults. A saved model that no longer appears in the catalogue remains visible as Unavailable so that Jaidu does not silently submit a different model. Choose an available model if the host can no longer run the saved one.

​

Runtime Resources

​

Choosing Runtime Resources

Use the runtime resource selector in the chat input to choose which skill and MCP server definitions are selected for the next Codex turn. You can use the workspace default, choose an organisation runtime preset, or customise the resources for one follow-up. The selection applies to direct sends, queued follow-ups, and message retries. It does not mutate the saved workspace runtime snapshot. Runtime presets choose resource access only:

• skills exposed by the selected host executor;
• MCP servers exposed by the selected host executor;
• default, preset, or custom resource modes.

Runtime presets do not change sandbox, approval, reasoning, prompt, command, or environment settings. Use the separate model selector to choose a model and reasoning level for a turn. That choice is sent with the turn and is not stored in the runtime preset. When you change the selected skills before sending a message, the changed skill set applies to that next workspace turn. For a resumed thread, Jaidu refreshes the thread’s skill inventory before Codex resumes so older preset turns do not intentionally carry deselected skills forward. If that refresh cannot be done safely, the send fails with a runtime cleanup error instead of continuing with stale skill instructions. The selector shows the selected resources and verifies them against the current host inventory. Its Selectable status means the resource can be passed to the next Codex turn; it does not mean MCP tools are already available to the agent. MCP servers still have a second launch-time step: Codex must start the selected server and the server must advertise callable tools. /mcp output is labeled when it is showing base host MCP status instead of the per-turn runtime preset.

​

Approval Workflow

Codex can ask for your approval before making changes. This gives you control over what gets implemented.

​

Why Approvals Exist

• Prevent unwanted changes - Review the plan before code is written
• Catch misunderstandings early - Ensure Codex understood your request
• Guide the approach - Steer Codex toward your preferred solution

​

Reviewing Plans

When Codex needs approval, you’ll see an approval card in the chat:

1. Read the plan summary - What Codex intends to do
2. Expand for details - Click to see the full plan with specific files and changes
3. Choose an action:
   • Approve - Codex proceeds with the plan
   • Request Changes - Provide feedback for revision

​

Requesting Changes

If the plan isn’t quite right:

1. Type your feedback in the chat input explaining what should be different
2. Click Request Changes
3. Codex revises the plan based on your feedback
4. Review the new plan and approve or request more changes

​

Disabling Approvals

If you want Codex to work autonomously without asking for approval:

1. Ask an administrator to review the host’s Advanced agent profile.
2. Confirm the workspace runtime preset still exposes only the skills and MCP servers needed for the task.

​

Editing Messages

You can edit and resend previous messages:

​

Status Indicators

​

Token Usage

The context gauge shows how much of the Codex context window is used. Understanding this helps you get better results.

​

What are Tokens?

Tokens are how AI models measure text. Roughly:

• 1 token ≈ 4 characters or ¾ of a word
• A 200K context window can hold about 150,000 words

The context window includes everything Codex has in conversation context: your messages, its responses, code it has read, and file contents.

​

Why Token Usage Matters

When usage is high:

• Codex may lose earlier parts of the conversation
• Responses may become less accurate
• Codex might re-read files it already read
• Complex reasoning may suffer

​

Usage Levels

​

Tasks Progress

When Codex breaks down work into tasks, a progress indicator appears in the chat toolbar:

• Progress bar showing completion percentage
• Task count (e.g., “5/5 completed”)
• Individual tasks with checkmarks when complete

Click the indicator to expand the full task list showing each step Codex is working through.

​

File Changes

The chat shows file modification summaries:

• Green numbers indicate lines added
• Red numbers indicate lines removed
• Click to view the file in the Changes panel

​

Tool Outputs

​

Command Execution

When Codex runs commands, you’ll see:

• Terminal icon for bash commands
• Exit code for completed commands
• Fix Script button if a command fails

Click to view full output in the logs panel.

​

Search Results

Search operations show summarised results. Click to expand and see full search output.

​

Review Comments

When you add inline comments in the Changes panel:

1. A banner appears showing the comment count
2. Comments are automatically included when you send a message
3. Click Clear to remove pending comments

​

Keyboard Shortcuts

​

Related Documentation

• Slash Commands - Quick actions with slash commands
• Threads - Managing multiple conversation threads
• Interface Guide - Overview of the workspace layout
• Command Bar - Quick actions and shortcuts