> ## Documentation Index > Fetch the complete documentation index at: https://docs.runchat.com/llms.txt > Use this file to discover all available pages before exploring further. # Agent Capabilities > Reference for every tool the chat agent and Agent node can use: read the canvas, create and connect nodes, edit code, run workflows, and search the web. The Runchat agent has access to a set of built-in tools for interacting with your canvas, editing code, loading skills, and more. The tools available depend on the current [permission level](/concepts/agents/chat#permission-levels). In plain language, the agent can: * Read everything on your canvas, including node configurations and error messages * Create and connect new nodes, organize layouts, and update parameters * Edit code in Code nodes with surgical replacements rather than full rewrites * Run nodes and execute workflows on your behalf (Run permission only) * Search, inspect and run published tools (both from the community and your own private tools) * Load domain-specific skills before writing code * Create, update and publish workflows (when used from the dashboard, REST API or MCP) The rest of this page lists the underlying tools by category. ## Canvas Tools ### Reading the canvas These tools are available at all permission levels: * **Get canvas** (`get_canvas`), see all nodes and edges on the canvas. For large canvases, paginate with the offset parameter. * **Read nodes** (`read_nodes`), inspect the full state of specific nodes by ID, including inputs, outputs and configuration. ### Creating and editing nodes These tools require **Edit** or **Run** permission: * **Create node**, add new nodes to the canvas. The agent can create all node types: Agent (`promptNode`), Create (`imageNode`), Code (`codeNode`), Input (`inputNode`), Note (`noteNode`), and Tool (`runChatNode`). * **Connect nodes**, wire data between nodes by specifying source and target handles. * **Update node**, modify an existing node's label, model, or input values. * **Delete nodes**, remove nodes from the canvas. * **Delete edges**, disconnect nodes without deleting them. * **Organize nodes**, auto-layout nodes after creating or modifying a workflow. ### Running nodes Requires **Run** permission: * **Run nodes**, execute nodes in dependency order with their current inputs and connections. ## Code File Tools Every Code node — regardless of mode — exposes a uniform file model. Snippet modes (JavaScript, HTML, Rhino/Blender Python, Revit C#) have a single virtual entry file (`index.js`, `index.html`, `index.py`, `index.cs`). Worker mode has `index.ts` plus optional additional files like `package.json` or helper modules. The same four tools work across all modes: * **Read files**, read content from one or more files in a code node with line numbers. Omit `files` to read everything (recommended when you don't yet know paths). Available at all permission levels. * **Read status**, check error and status messages for debugging. Available at all permission levels. These require **Edit** or **Run** permission: * **Edit file**, targeted edit (pass `old_text` + `new_text`) or full overwrite (pass only `new_text`). * **Create files**, add new files to a worker-mode bundle. Worker mode only. * **Delete files**, remove non-entry files from a worker-mode bundle. Worker mode only. ## Media Tools * **Get model parameters**, fetch the full parameter details for a generative model before creating an `imageNode`. This tells the agent what inputs the model accepts (e.g. prompt, image\_url, resolution). * **View image**, fetch and display images from URLs in tool results. ## Custom Tools * **Search Tools**, search for public tools published by the community and private tools published by your team. * **Inspect Tool**, view the workflow of a specific tool. * **Execute Tool**, with run permissions, run the tool. Common patterns: * **Two-step research for better images**, ask the agent to "search first, then pull images from the projects you find" rather than grabbing search thumbnails. You'll get higher-resolution imagery straight from the source. * **Group on the canvas**, once images are nodes on the canvas, the agent can read them and group by palette, narrow to a brief, or compare two. * **Pick a format**, ask for **Notes** when each item has mixed content (image, description, link); ask for **Tables** when you want to compare consistent columns across items. * **Deep dive a result**, pick one project from the search and ask the agent to dig further. It can follow links to find architect statements, materials, press coverage, then consolidate everything onto the canvas.