How-tos
Task-oriented recipes. Each page is structured as problem -> minimal YAML -> deploy + verify, with the live test command included. Where a recipe touches a module, the page links to the full module reference instead of repeating it.
Standalone how-tos
| Recipe | Page |
|---|---|
| Build a RAG bot over a folder of markdowns | Build a RAG bot |
| Attach files to a chat session (paperclip menu) | Attach files to chat |
| Add a new module to the daemon (contributor doc) | Add a module |
| Install an MCP server (Hub catalog or custom) | Install an MCP server |
Linked recipes
Topics where the canonical reference page is already exhaustive; the howto would just be a copy. Follow the links instead.
| Recipe | Where to look |
|---|---|
| Add a new LLM provider | Agents - Provider examples |
| Secure shell access (allowed/blocked commands) | Shell module, OS Sandbox |
| Debug an app (logs, dev CLI, tracing) | Dev CLI, Observability |
| Use the credentials vault | Credentials |
| Migrate from legacy v1 YAML | Language - migration table |
| Deploy to production | Production deployment, Deployment |
| Build a multi-agent team | Tutorial - Multi-agent team |
When to write a howto
Write a howto when:
- The task crosses 3+ pages of the reference and you want a single copy-pasteable answer.
- The task has a non-obvious gotcha (an order-of-operations, a credential scope, a sandbox interaction).
- A user asks the same question twice on the issue tracker.
A howto is NOT the place for new conceptual material. If the recipe needs a paragraph of "why", that paragraph belongs in Concepts and the howto links to it.