Kilncast

kilncast

Thin typed wrapper over Firestore. It types and coerces, it does not validate: the schema is used for type inference only and is never executed, so there is no runtime validation. Underneath it is plain Firestore; every handle exposes .ref to drop to the raw SDK.

Setup

# server
npm install kilncast firebase-admin
# web
npm install kilncast firebase

Three entrypoints:

• kilncast: neutral. Schema definition, sentinels, types, KilncastError. Imports no Firebase, safe in shared modules and frontend bundles.
• kilncast/admin: createDatabase(getFirestore()) over firebase-admin, plus typed ref helpers. Re-exports the whole neutral core.
• kilncast/web: the same over the modular firebase SDK.

Quick start

import { collection, increment } from "kilncast";
// or "kilncast/web"
import { createDatabase } from "kilncast/admin";
import { getFirestore } from "firebase-admin/firestore";
import { z } from "zod";

// Plain, unbound definition, define once in a shared module importing only "kilncast"
const posts = collection(
  "posts",
  z.object({ title: z.string(), likes: z.number(), createdAt: z.date() }),
);

const db = createDatabase(getFirestore());

await db
  .collection(posts)
  .set("hello", { title: "Hello", likes: 0, createdAt: new Date() });
// Doc | null, Doc = T & { id }
const post = await db.collection(posts).get("hello");
await db.collection(posts).update("hello", { likes: increment(1) });
const popular = await db
  .collection(posts)
  .where("likes", ">=", 10)
  .orderBy("likes", "desc")
  .limit(10)
  .get();

The schema is any Standard Schema validator (Zod, Valibot, ArkType). It describes one document's fields, excluding the id.

Core rules

1. Ids are path-only. Never declare an id field in a schema (compile error). Reads merge id in flat; writes exclude it. add(data) returns the generated id.
2. Missing reads are null. get() returns Doc | null; there is no .exists snapshot to check (use .doc(id).exists() for a boolean).
3. Values coerce at the boundary. Write a Date, read a Date (stored as Timestamp); write a Uint8Array, read a Uint8Array (stored as SDK bytes). Deep through maps and arrays.
4. Sentinels are kilncast's own. Import serverTimestamp, increment, arrayUnion, arrayRemove, deleteField from kilncast, never FieldValue from an SDK. Each is type-constrained to the fields it fits (increment on numbers, deleteField on optional fields only).
5. Queries are typed to the schema. where / orderBy take typed dotted paths into nested maps ("customer.address.city"); a misspelt field or wrong value type is a compile error. Target the id with documentId() from kilncast.
6. Dotted-path keys work in update only. update("o1", { "totals.items": increment(1) }) touches one nested field. In a merge set a dotted key is a literal field name, so nest an object instead.
7. Transaction and batch writes take no await. Reads in a transaction are async; writes return void and buffer until commit. Batches are write-only (no get, no add) and apply atomically on .commit().
8. No runtime validation. A document drifted from its schema comes back typed as valid. Where drift matters, run the schema on the result yourself.
9. SDK semantics win. kilncast never wraps Firestore, network, or permission errors; KilncastError covers only kilncast's own failures. Anything kilncast does not wrap is reachable via .ref.

Common operations

const col = db.collection(posts);

// Reads
// Doc | null
await col.get("hello");
// boolean
await col.doc("hello").exists();

// Writes, also on col.doc("hello")
await col.set("hello", { title: "Hi", likes: 0, createdAt: new Date() });
await col.set("hello", { likes: 1 }, { merge: true });
await col.update("hello", { likes: increment(1) });
// returns the new id
const id = await col.add({ title: "New", likes: 0, createdAt: new Date() });
await col.delete(id);

// Queries, immutable builder, .get() alias .list()
await col.where("likes", ">=", 10).orderBy("likes", "desc").limit(5).get();
await col.count();
// 0 over empty match
await col.sum("likes");
// null over empty match
await col.average("likes");
// returns unsubscribe fn
const unsub = col.onSnapshot((docs) => {});

// Subcollection, same def reusable at any depth
db.collection(posts).doc("hello").collection(comments);

// Collection group, every collection with that name
await db.collectionGroup(comments).orderBy("createdAt").get();

// Transaction, reads async, writes sync
await db.runTransaction(async (tx) => {
  const post = await tx.collection(posts).get("hello");
  if (post) tx.collection(posts).update("hello", { likes: post.likes + 1 });
});

// Batch, write-only, atomic on commit
const batch = db.batch();
batch.collection(posts).update("hello", { likes: increment(1) });
await batch.commit();

Full reference

See [REFERENCE.md](REFERENCE.md) for queries and cursors, sentinels and nested updates, live updates, transactions and batches in detail, raw fields and full-precision timestamps, bytes, neutral value types, the .ref escape hatch, and gotchas.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: mwpryer
• Source: mwpryer/kilncast
• License: MIT
• Homepage: https://www.npmjs.com/package/kilncast

Install and usage instructions live in the source repository linked above.