# Coding agents SnoutData can run **your own coding agent inside the app**, in a panel, in your workspace folder, with your database connections handed to it the moment it starts. That last part is the point. Normally, giving an agent access to a database means putting a connection string in a `.env` file or a repo and hoping it stays there. Here the app is already connected, and it lends the agent that access without ever handing over the credentials themselves. ## Which agents | Agent | Command | |---|---| | Claude Code | `claude` | | Codex | `codex` | | opencode | `opencode` | An agent gets an icon in the activity bar when it is **installed on this computer** and you have not switched it off. If you have none of them installed, nothing appears, and nothing nags you about it. ## Open one Press `Ctrl+``` (backtick), click the agent's activity-bar icon, or run **Agent Console: Show/Hide** from the command palette. The agent starts in your current workspace folder, so it sees the files you are working on. If the agent is not installed, the panel tells you where to get it rather than failing silently. ## Closing the panel does not kill the agent Closing the console **detaches** the session instead of ending it. A long job keeps running while the panel is away, and reopening the panel reattaches to it. ## What it can do Two separate things, and they are worth reading in order: - [What the agent can do](https://docs.snoutdata.com/agents/what-it-can-do): querying your databases, driving the app, and drawing real interface into it. - [Approvals and audit](https://docs.snoutdata.com/agents/approvals): how you stay in control of everything it changes. If you would rather keep your agent where it already lives, outside SnoutData, use [agent access over MCP](https://docs.snoutdata.com/agent/mcp) instead. It is the same broker and the same rules.