Agent Panel

The Agent Panel is NexusIDE's primary interface for interacting with NexusCore. It provides a chat-based workflow with multiple agent modes, tool approval controls, and real-time progress tracking.

Opening the Panel

• Click the NexusCore icon in the activity bar (left sidebar)
• Or use the keyboard shortcut Ctrl+Shift+N / Cmd+Shift+N
• Or open the command palette and run NexusCore: Open Agent Panel

Agent Modes

The chat panel input area includes a mode selector with three options:

Ask Mode (Default)

Single-turn Q&A. The agent answers your question without making any file changes.

• Best for: Quick questions, code explanations, debugging help
• No tools are used — the agent responds based on context only
• Default mode for new conversations

Edit Mode

The agent can read and edit files in your workspace, but each change requires your approval.

• Best for: Focused code changes, bug fixes, small refactors
• The agent proposes changes as inline diffs
• You review and accept or reject each change individually

Agent Mode (Autonomous)

Multi-step execution where the agent plans, uses tools, and works through the task autonomously.

• Best for: Complex tasks, multi-file refactoring, project scaffolding
• A step budget indicator shows the maximum tool calls allowed (default: 25, configurable)
• A progress indicator shows the current step count during execution
• Click Stop to interrupt execution at any time
• When complete, the agent shows a summary of all actions taken with accept/reject controls for each change

TIP

The mode selector state persists across chat sessions. You don't need to re-select your preferred mode each time.

Swarm Mode

For complex tasks that benefit from parallelization, toggle Swarm alongside Agent mode:

1. Enable the Swarm toggle in the chat panel
2. Send your task — the agent decomposes it into sub-tasks
3. Parallel sub-agents execute simultaneously
4. Open the Swarm Visualizer panel from the sidebar to monitor progress

The swarm visualizer shows:

• Parent task and decomposed sub-tasks as a tree
• Each sub-agent's status (queued, running, completed, failed)
• The model assigned to each sub-agent
• Elapsed time per sub-agent

You can cancel individual sub-agents or the entire swarm from the visualizer.

Swarm limits:

• Maximum concurrent sub-agents: 5 (configurable)
• Per-child timeout: 120 seconds (configurable)

Tool Approval

The agent uses tools to interact with your workspace. Each tool has an approval policy:

Policy	Behavior	Default For
Auto-approve	Runs without asking	Read-only tools: file_read, file_list, text_search
Ask	Requires your approval each time	Write tools: file_write, file_delete, terminal_run, git_commit
Deny	Never allowed to run	None (user-configured)

When the agent requests a tool with Ask policy, an inline prompt appears in the chat:

NexusCore wants to run: file_write("src/utils.ts")
[Allow] [Allow for Session] [Always Allow] [Deny]

• Allow — Run this one time
• Allow for Session — Auto-approve this tool for the rest of the session
• Always Allow — Permanently change the tool's policy to auto-approve
• Deny — Block this execution

Configure tool policies in Settings → NexusCore → Trust & Approvals.

Status Indicators

The status bar at the bottom of the IDE shows the agent's current state:

Status	Meaning
NexusCore · Ready	Agent is idle and ready for input
NexusCore · Working	Agent is processing your request
NexusCore · Free	Not signed in — Core tier features available
NexusCore · Pro	Signed in with Pro tier
NexusCore · Studio	Signed in with Studio tier

The active model name (e.g., qwen2.5-coder-32b) appears next to the status icon. Click it to switch models.

Model Badges

When hybrid routing is active (different models for different steps), each response segment shows a subtle model badge indicating which model generated it. This helps you understand how routing decisions are made and tune your configuration.

Tips

• Use Ask mode for quick questions — it's faster because no tools are invoked
• Use Edit mode when you want to review each change before it's applied
• Use Agent mode for tasks that require multiple steps and tool use
• Enable Swarm for large tasks that can be parallelized (e.g., "add tests to all modules")
• Check the Output panel (channel: "NexusCore Routing") to see model routing decisions