Your first agent
Ironspire lets you run multiple AI agents simultaneously, each with their own tools, context, and personality. Here is how to get your first one running.
Adding an agent
- Click the Add Agent button in the left sidebar (or press the keyboard shortcut)
- The Add Agent Modal opens, showing available presets
- Pick a preset that matches your task (or start with Developer for general coding work)
- Give your agent a name (optional; Ironspire generates one if you skip this)
- Click Create
Your agent appears in the sidebar and is ready to chat.
Understanding presets
Presets are pre-configured agent templates. Each preset comes with:
- A system prompt tailored to the role
- MCP tools relevant to the task (filesystem, web search, code analysis, etc.)
- A colour and icon for quick visual identification
Here is a quick overview of the 9 built-in presets:
| Preset | Role | Key tools |
|---|---|---|
| Project Manager | Project management, planning | Task tracking, documentation |
| Developer | Code writing, debugging | Filesystem, code tools |
| Data Analyst | Data analysis, reporting | File tools, code execution |
| UI/UX Designer | UI/UX guidance, design specs | Screenshot tools, filesystem |
| Marketing Specialist | Marketing strategy, campaigns | Web search, analytics tools |
| Creative Director | Creative direction, brand | Visual tools, web search |
| QA Engineer | Testing, bug reporting | Code tools, filesystem |
| DevOps Engineer | Infrastructure, deployment | GitHub, Cloudflare, Sentry, Filesystem |
| Technical Writer | Documentation, tech writing | File writing, web search |
Each agent gets its own conversation thread, working directory, and tool permissions. They do not interfere with each other.
Start chatting
Once your agent is created, click on it in the sidebar to open its chat panel. Type your message and press Enter to send.
The agent will respond using the AI model configured in your settings. You can see real-time token usage, cost tracking, and context window status in the session HUD above the chat.
Permission and interaction modes
Ironspire separates control into permission modes and interaction modes:
Permission modes (2)
- default: the agent operates within standard permission boundaries
- full_access: bypasses all permission checks (maps to
bypassPermissionsinternally)
Interaction modes (3)
- normal: the agent asks before using risky tools (default)
- plan: the agent creates a plan but does not execute it
- auto-accept: the agent runs tools without asking (for trusted workflows)
Auto-accept and full_access modes let agents modify files and run commands without confirmation. Use them only when you understand what the agent will do.