Agent
Agent is Velork's assistant for design-aware software work. It can explain your repo, edit files, run shell commands, use MCP tools, and implement UI from canvas selections. Open chat from the right auxiliary bar and press Cmd + L to focus the input.
How Agent works
An agent session combines three components:
- Instructions — your prompt, the selected mode, and any active blueprint
- Tools — file edits, search, terminal, MCP, canvas and inspiration tools
- Model — the model from your connected provider, chosen in chat or Settings → Models
Velork attaches context automatically: open files, selected canvas frames, and the active task when relevant.
Tools
Agent uses tools to gather information and apply changes. There is no fixed limit on tool calls per task — each run continues until the job is done or you stop it.
File and code tools
Search the repo, read files, and propose edits. Diffs appear inline for review before you accept.
Terminal
Run shell commands and inspect output. Configure bash permissions under Settings → Tools & MCP.
Canvas and inspiration
Read the design tree from selected frames. In Inspiration mode, capture live sites for reference before switching to Develop to implement.
MCP
Call external tools — GitHub, Linear, Context7, or custom servers. See MCP servers.
Tool permissions
When Agent requests a tool, Velork shows a prompt:
- Run Tool — Cmd + Enter
- Skip Tool — Escape
Set default policies in Settings → Tools & MCP. Autonomous agent mode (Cmd + Shift + A) reduces prompts but still respects hard blocks you configure.
Modes and blueprints
Modes tune behavior — Develop for code, Plan for exploration, Test for verification, and more. See Modes.
Blueprints are reusable prompt templates in the chat input. Free plans include a small library; Pro and Team are unlimited.
Next steps
Chat
Prompting patterns and session management
Modes
When to use Develop, Plan, Test, and others
Tasks
Queue agent work from the task board
AI providers
Connect Copilot, Codex, Claude Code, or Cursor