Copilot

Copilot is an AI-powered assistant that works across the entire TradingGoose workspace — not just workflows. It can manage workflows, MCP servers, skills, custom tools, knowledge bases, deployments, variables, and more. In the desktop workspace, open or close its persistent panel with the Copilot control in the global sidebar. The panel is not available in the mobile workspace layout.

This page covers the built-in Studio assistant. To work on TradingGoose from your own coding agent, see Copilot MCP for capabilities, installation, and authentication.

Capabilities

Copilot has access to a comprehensive set of tools spanning the entire platform:

Workflow Management

• Edit workflows — add, remove, or reconfigure blocks and connections. Edits stream in real time as inline previews before being committed.
• Deploy workflows — check deployment status and trigger new deployments.
• Run workflows — execute workflows and inspect results.
• Manage variables — get and set global workflow variables.
• View console — retrieve execution logs and streaming output.

Workspace Tools

• Manage MCP servers — create, update, and configure MCP server connections.
• Manage skills — create, edit, and rename reusable skill definitions.
• Manage custom tools — create and edit custom tool schemas and code.
• Knowledge base — query and reference knowledge base documents.

General

• Explain — answer questions about TradingGoose concepts or any workflow in your workspace.
• Plan — for larger tasks, Copilot generates a todo list that tracks progress as it works through each step.
• Reference context — use the @ mention menu to pull in previous chats, workspace entities, specific blocks, docs, or execution logs.
• File attachments — attach files directly to a message for additional context.

Access Levels

The access-level toggle in the input bar controls how much autonomy Copilot has.

Supported Models

Copilot supports exactly six OpenRouter models. Studio sends each canonical OpenRouter model slug unchanged, and you can switch the active model from the input bar.

Anthropic -- anthropic/claude-fable-5, anthropic/claude-opus-5

DeepSeek -- deepseek/deepseek-v4-pro

OpenAI -- openai/gpt-5.6-terra, openai/gpt-5.6-sol

xAI -- x-ai/grok-4.6

Setup

The Copilot service API key above connects the built-in assistant to managed inference. Copilot MCP uses a personal TradingGoose API key obtained through its own browser approval flow.

Chat Management

Each Copilot session is scoped to the authenticated user and workspace. The header dropdown lists all past chats grouped by recency (Today, Yesterday, This Week, Last Week, Older). From the header you can:

• Switch chats -- click any previous conversation to resume it.
• Start a new chat -- click the + button in the header to create a fresh session.
• Delete a chat -- hover a chat entry and click the trash icon. A confirmation dialog prevents accidental deletion.

Chat state persists while you navigate the workspace or collapse and reopen the panel. Copilot follows the current page and selection: for example, a knowledge detail, selected execution log, or monitor. On the dashboard, the current context identifies only the active layout. Copilot can read that layout through its dashboard tools when the task requires more detail.

Editor Integration

Copilot is context-aware. The current page publishes only the selected object's identity, such as the active dashboard layout, knowledge base, execution log, or monitor. Use the @ mention menu when you want to add other explicit context to a message.

When Copilot streams a response that includes tool calls (for example adding a block, creating a skill, or configuring an MCP server), each action is rendered as an inline preview. In Limited mode you approve or reject each change; in Full mode changes are applied automatically.

You can edit an earlier prompt to truncate and regenerate the messages that follow it, but only when no accepted live mutation follows that prompt. Editing chat history does not undo changes already applied to the workspace; reverse an applied action by changing the affected entity again. Messages after an editable point are dimmed to indicate that they will be regenerated.

Billing

Copilot completion charges use the authoritative total cost reported by OpenRouter, followed by the Copilot cost multiplier for your plan. This keeps final billing aligned with the actual routed request, including applicable cached-token, reasoning, media, search, long-context, and provider-specific charges.