> ## Documentation Index
> Fetch the complete documentation index at: https://xura.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# Plan from description

> Let Xura generate a Flow graph from a natural-language description.

The planner reads your description + your available tools + your live data schema, and returns a `{ nodes, edges }` graph ready to run.

## When you'll use it

* Empty flow with only a Trigger — you want a first draft.
* Blank slate — you know what to do but don't want to click through 5 node types.
* Priority-seeded flow — clicking an "Act" priority on the home page lands you here with the description prefilled.

## How to use it

1. Open a flow (or create a new one).
2. Type the description — what to check, what data to look at, what action to take.
3. Click **Plan from description** (top-right of the toolbar, green pill).
4. Wait \~5-15 seconds. The canvas repopulates with the generated graph.

## What makes a good description

The planner is only as good as your prompt. Bad ones get rejected client-side (under 20 chars or under 4 words). Good ones give the planner enough hooks.

**Bad** (rejected):

* `"HR stuff"`
* `"Sales pipeline"`
* `"Fix things"`

**Good** (planner has enough to work with):

* `"Every Monday, find deals >$50K stuck in negotiation for 14+ days and email the owner a reminder."`
* `"When any project milestone slips more than 3 days, escalate to the project sponsor with a delay summary."`
* `"Check burnout risk in R&D — flag anyone above 100% utilization and suggest reassignments."`

Structure that works: **trigger + filter + condition + action**.

## What the planner sees

Under the hood, Xura sends the LLM:

* Your description.
* Your **connected data entities** (names + fields, so it can't hallucinate).
* Your **available tools** (built-in, MCP, integrations — only the ones you've connected).
* Your **stored agents** (if you have any).

Then the LLM produces a JSON graph and Xura renders it.

## The disable-once-generated behavior

After a successful plan, the button greys out. It re-enables the moment you edit the description — this prevents accidentally regenerating the same graph and wasting tokens.

## If the plan is wrong

* **Edit nodes directly** in the graph editor. Each node has an Inspector panel for query text, tool args, etc.
* **Refine the description** and re-plan — the button re-enables when the text changes.
* **Delete + re-plan** — remove the generated nodes, tighten the description, click Plan again.