Agents
Set up an agent once, then give it a model, instructions, files, skills, memory, and tools.
An agent is the part of Blazing Agents you configure. You choose its model, write its instructions, and decide what it can use: files, skills, memory, and tools. Your app then calls it by ID for chat, one-off answers, or background work, and every call runs with that setup.
If you have not built one yet, start with Run your first agent.
What you can set up
| Page | What it lets you do |
|---|---|
| Agents | Create an agent and choose what it controls. |
| Providers and models | Connect your model account and pick the model. |
| Prompts | Save message templates and fill them in per call. |
| Versions and lifecycle | Roll back changes, pin a version, and pause an agent. |
| Skills | Teach workflows the agent loads only when needed. |
| Memory | Remember facts about each user across sessions. |
| Workspaces | Give the agent files and a shell that persist. |
| Artifacts | Hand files from a conversation back to your app. |
| Output | Stream text or get JSON that matches a schema. |
| Tools | Turn on built-in tools, connect MCP servers, and require approvals. |
How the pieces fit
Each call to an agent is a turn. On every turn, Blazing Agents reads the agent's current settings, or a version you pinned, and sends the conversation to its model. During the turn the agent can load a skill, read or save memory, work with files in its workspace, and call tools. The answer streams back as text or arrives as structured JSON.
Turns that belong to a conversation live in a session. Scheduled and background turns run as tasks.
Next
- Create an agent and see everything it controls.
- Connect a provider if your agent has no model yet.
- Use the TypeScript SDK or Python SDK for exact methods and types.