Skip to main content
Luminy’s agent is not a passive text generator — it actively works inside your project using a set of built-in tools. When you ask it to implement a feature, fix a bug, or run your tests, the agent plans a sequence of tool calls: reading the relevant files, making precise edits, running commands to verify the result, and checking in with you when it needs clarification. Every action is visible in the chat so you always know what it did and why.

​
How tools appear in chat

Each time the agent calls a tool, a tool badge appears inline in the conversation. The badge shows the tool name and a short summary — for example, Reading src/main.rs or Running npm test. Click the expand arrow on any badge to see the full input the agent sent and the output it received. When the agent calls several tools at once (a common pattern when it reads multiple files in parallel), they appear as a grouped batch badge. Expand the group to inspect each individual call.
You can press the Stop button at any time to cancel the current agentic run. The agent stops cleanly after the in-flight tool call finishes.

​
Tool categories

File System

Read, write, and edit files in your project. Search by glob pattern, grep file contents by regex, and look up function or class definitions by name.

Shell & Git

Run terminal commands and git operations directly in your project. Output streams live into the chat as commands execute.

MCP Servers

Connect external Model Context Protocol servers to add tools like web search, database access, cloud APIs, and custom integrations.

Vision

Analyze screenshots and images you attach to messages. The agent can describe what it sees, spot layout defects, and act on visual information.

Todo List

For multi-step tasks the agent creates and maintains a task list in the chat, marking items in-progress and done as it works through them.

Clarify

When the agent hits a genuine ambiguity it can’t resolve from the code, it pauses and asks you a structured set of questions rather than guessing.

​
Permissions

Some tools — especially shell commands — are sensitive. Before the agent can run them, it may ask for your approval. You can configure whether approval is required per-run, always granted, or always denied in Settings → Permissions.
Enable per-command approval in Settings → Permissions if you want to review every shell command before it executes. This is especially useful when working in a production codebase.
See the Permissions page for a full explanation of the approval modes.

​
Stopping the agent

You can interrupt any agentic run by clicking the Stop button that appears in the chat while the agent is working. The agent halts after the current tool call completes and returns control to you.