Agent Overview
Vyre Agent is an autonomous coding assistant that reads your codebase, plans changes, uses tools, and writes code — all in response to a single natural-language instruction.
Unlike a chat assistant that only suggests code, Vyre Agent directly edits files, runs terminal commands, reads documentation, and browses the web to complete tasks end-to-end.
How Agent works
Agent operates in a loop. When you give it a task:
- It reads — Agent searches your codebase, reads relevant files, and builds a mental model of your project.
- It plans — For complex changes, Agent identifies the files to touch and the order of operations.
- It acts — Agent calls tools: editing files, running commands, browsing URLs, or asking you a clarifying question.
- It verifies — Agent reads its own output and catches mistakes before reporting back.
This loop continues until the task is done or Agent needs your input.
What Agent can do
Read & Search Files
Agent reads files, searches by keyword or semantics, and navigates your project structure to build context.
Edit Code
Agent creates new files, edits existing ones, and can perform targeted in-place replacements across many files.
Run Terminal Commands
Agent can run build scripts, test runners, package install commands, and any other shell commands.
Browse the Web
Agent can read documentation pages, GitHub issues, and Stack Overflow to gather context from the internet.
Read Images
Attach screenshots or design mockups and Agent will implement the UI directly from your visual reference.
Ask Questions
When requirements are ambiguous, Agent pauses and asks clarifying questions rather than guessing.
Agent vs Ask
When to use Agent
Use Agent when you want Vyre to do work: write code, edit files, run commands, and complete a task. Agent is autonomous — it will keep working until the task is done.
Best for:
- Implementing new features
- Refactoring code across multiple files
- Fixing bugs end-to-end
- Running tests and interpreting results
When to use Ask
Use Ask when you want Vyre to explain something or answer a question without modifying your code. Ask is read-only.
Best for:
- Understanding unfamiliar code
- Asking about architecture decisions
- Getting a second opinion
- Researching how a library works
Checkpoints
Every time Agent finishes a meaningful chunk of work, Vyre automatically creates a checkpoint — a snapshot of your workspace state. If Agent's output isn't what you wanted, you can restore any previous checkpoint from the agent panel without losing your work.
Use checkpoints as a safety net for large refactors. Let Agent work freely, then review the diff and restore to an earlier checkpoint if needed.