Quick Start

This page walks through a minimal end-to-end query: define a fetcher, subscribe to it with use_query, read the status and data, and render. Then a tiny use_mutation example so you can see the full loop.

It assumes you have already installed gpui-query and set up a QueryClient global.

The mental model

gpui-query is built around three pieces:

A fetcher: an async function that knows how to load some data. gpui-query owns everything else: caching, deduplication, retries.
use_query: a hook you call from a view’s constructor (not in render). It fetches the first time, subscribes the view to the result, and returns an Entity you read during render.
The QueryClient: a global registry that shares results across views so the same key is only ever fetched once.

Data is cached under a QueryKey. Two views asking for the same key read from the same cached entity.

Your first query

1. Model your data and error

Both the data type T and the error type E must be Clone + Send + Sync + 'static, and E must be Debug. A cheap stand-in for the error is the crate’s QueryError, but you can use your own.

#[derive(Clone, Debug)]
pub struct Todo {
    pub id: u32,
    pub title: String,
    pub done: bool,
}

2. Define a fetcher

A fetcher is a Fn() -> Future<Output = Result<T, E>>. It takes no arguments and returns the data or an error. The real networking is up to you (reqwest, an LSP transport, a local file). Here we return a fixed list so the example compiles without a server.

use gpui_query::QueryError;

async fn fetch_todos() -> Result<Vec<Todo>, QueryError> {
    // Replace this with a real request.
    Ok(vec![
        Todo { id: 1, title: "Write the docs".into(), done: false },
        Todo { id: 2, title: "Ship it".into(), done: true },
    ])
}

Call use_query in your view’s new. It returns a tuple: the resource entity (store it on the view to read during render) and a Subscription (store it too, because dropping it stops the view from re-rendering on changes).

use gpui_query::hook::use_query;
use gpui_query::{CachePolicy, QueryError, QueryResource, core::RequestPolicy};

pub struct TodosView {
    todos: gpui::Entity<QueryResource<Vec<Todo>, QueryError>>,
    _sub: gpui::Subscription,
}

impl TodosView {
    pub fn new(cx: &mut gpui::Context<Self>) -> Self {
        let (todos, sub) = use_query(
            "todos",                                  // cache key
            CachePolicy::Ttl { ttl_ms: 60_000 },      // fresh for 1 minute
            RequestPolicy::LatestWins,                // supersede in-flight fetches
            || async { fetch_todos().await },         // fetcher: Fn() -> Fut
            cx,
        );
        Self { todos, _sub: sub }
    }
}

use_query starts a fetch the first time the resource is seen, and reuses the cached entity on every subsequent access for the same key. Pass a fresh QueryKey (QueryKey::from(["todos"]) for a multi-segment key) when you want hierarchical keys for invalidation.

4. Read status and data, then render

A QueryResource<T, E> is a normal GPUI entity. Read it with entity.read(cx) and branch on status():

impl gpui::Render for TodosView {
    fn render(&mut self, _window: &mut gpui::Window, cx: &mut gpui::Context<Self>) -> impl gpui::IntoElement {
        let r = self.todos.read(cx);

        match r.status() {
            gpui_query::QueryStatus::Idle
            | gpui_query::QueryStatus::LoadingEmpty => {
                // First load, no data yet.
                gpui::div().child("Loading…")
            }
            gpui_query::QueryStatus::Success => {
                let list = r.data().cloned().unwrap_or_default();
                gpui::div().children(
                    list.into_iter().map(|t| gpui::div().child(t.title)),
                )
            }
            gpui_query::QueryStatus::Failure => {
                let msg = r.error().map(|e| format!("{e:?}")).unwrap_or_default();
                gpui::div().child(format!("Failed: {msg}"))
            }
            _ => gpui::div(),
        }
    }
}

Useful predicates on the resource:

r.is_loading(): true during the first fetch or any refetch.
r.is_pending(): true when there is no data yet and a fetch is in flight.
r.data() / r.error(): Option<&T> / Option<&E>, borrowed for the duration of the read.

That is the whole query loop: define a fetcher, subscribe with use_query, branch on status, render. gpui-query handles caching, deduplication, and re-rendering on every status transition.

A tiny mutation

Mutations are imperative: you trigger them on demand (a button, a form submit). use_mutation creates a MutationResource<V, T, E> entity: V is the variables you pass in, T the success result, E the error. Trigger it with the mutate helper.

use gpui_query::hook::{use_mutation, mutate};
use gpui_query::{MutationResource, MutationStatus, QueryError};

#[derive(Clone)]
pub struct ToggleTodo { pub id: u32 }

pub struct TodoActions {
    toggle: gpui::Entity<MutationResource<ToggleTodo, Todo, QueryError>>,
}

impl TodoActions {
    pub fn new(cx: &mut gpui::Context<Self>) -> Self {
        // use_mutation returns the entity; turbofish fixes the (V, T, E, C) quadruple.
        let toggle = use_mutation::<ToggleTodo, Todo, QueryError, Self>(cx);
        Self { toggle }
    }

    pub fn on_toggle(&mut self, id: u32, cx: &mut gpui::Context<Self>) {
        mutate(
            &self.toggle,
            ToggleTodo { id },                  // variables
            |vars| async move {                 // mutator: Fn(V) -> Fut
                // Replace with a real PATCH/POST.
                Ok(Todo { id: vars.id, title: "Toggled".into(), done: true })
            },
            cx,
        );
    }
}

Read the mutation’s status the same way as a query:

# use gpui_query::{MutationResource, MutationStatus, QueryError};
# #[derive(Clone)] struct ToggleTodo { id: u32 }
# #[derive(Clone)] struct Todo { id: u32, title: String, done: bool }
# fn doc(m: &gpui::Entity<MutationResource<ToggleTodo, Todo, QueryError>>, cx: &gpui::App) {
match m.read(cx).status() {
    MutationStatus::Idle => { /* render a button */ }
    MutationStatus::Loading => { /* disable the button, show a spinner */ }
    MutationStatus::Success => { /* confirm */ }
    MutationStatus::Failure => { /* render a retry button */ }
}
# }

Mutations default to no retries. After a mutation that changes server state, invalidate the affected queries so they refetch:

# use gpui_query::client::QueryClient;
# use gpui_query::core::{QueryKey, QueryKeyFilter};
# fn doc(cx: &mut gpui::App) {
# let mut client = cx.global::<QueryClient>().clone();
let todos = QueryKey::from(["todos"]);
client.invalidate_queries(&QueryKeyFilter::Prefix(&todos), cx);
# }

Where to go next

You now have a query that caches and a mutation that triggers. From here:

Queries: the full use_query surface, the fetcher signal, and refetching.
Mutations: lifecycle callbacks and the mutate variants.
Caching: CachePolicy, request deduplication, and GC.
Query keys: hierarchical keys and invalidation scope.