# Have a useful conversation

Give Helm enough context to help, and verify the outcome.

Source: https://docs.minds.sh/docs/helm/conversations



Start with the result you want, then add the details that change the answer. You do not need to phrase a request as an API call.

## A useful request has three parts [#a-useful-request-has-three-parts]

| Part       | Example                                                       |
| ---------- | ------------------------------------------------------------- |
| Goal       | “Help me set up a Mind for support history.”                  |
| Scope      | “This workspace serves our customer support assistant.”       |
| Constraint | “Start on Free and choose a region close to our application.” |

For questions about existing information, identify the relevant project, time period, or source. Ask Helm to distinguish a stored fact from an inference. If a cited source is not available, treat the answer as guidance rather than verified recall.

## Use the composer [#use-the-composer]

Press **Enter** to send and **Shift+Enter** for a new line. The context control can insert a memory, keyspace, or file reference token. These tokens identify context in the draft; they are not an upload or a guarantee that a source has been retrieved.

## Before acting on a reply [#before-acting-on-a-reply]

Check the selected instance, the target resource, and the action shown in a card. For a completed operation, confirm the resulting state in the workspace or through a fresh API request. Do not infer that a request ran from conversational wording alone.

If you need a deterministic integration, use the [API reference](/docs/api) and handle its response explicitly.